Impacts of ciliary neurotrophic factor on the retinal transcriptome in a mouse model of photoreceptor degeneration

Ciliary neurotrophic factor (CNTF) has been tested in clinical trials for human retinal degeneration due to its potent neuroprotective effects in various animal models. To decipher CNTF-triggered molecular events in the degenerating retina, we performed high-throughput RNA sequencing analyses using the Rds/Prph2 (P216L) transgenic mouse as a preclinical model for retinitis pigmentosa. In the absence of CNTF treatment, transcriptome alterations were detected at the onset of rod degeneration compared with wild type mice, including reduction of key photoreceptor transcription factors Crx, Nrl, and rod phototransduction genes. Short-term CNTF treatments caused further declines of photoreceptor transcription factors accompanied by marked decreases of both rod- and cone-specific gene expression. In addition, CNTF triggered acute elevation of transcripts in the innate immune system and growth factor signaling. These immune responses were sustained after long-term CNTF exposures that also affected neuronal transmission and metabolism. Comparisons of transcriptomes also uncovered common pathways shared with other retinal degeneration models. Cross referencing bulk RNA-seq with single-cell RNA-seq data revealed the CNTF responsive cell types, including Müller glia, rod and cone photoreceptors, and bipolar cells. Together, these results demonstrate the influence of exogenous CNTF on the retinal transcriptome landscape and illuminate likely CNTF impacts in degenerating human retinas.

Retinal degeneration (RD) is known as an irreversible, progressive neurologic disorder caused by genetic mutations and/or environmental or pathological damage to the retina. One therapeutic strategy to attenuate vision loss in disease conditions is treating the retina with neuroprotective agents that promote neuronal viability. Ciliary neurotrophic factor (CNTF) has been shown to be a potent neuroprotective agent for various types of retinal degenerations, including those caused by gene mutations, light-induced damage, and physical injury 1 . Since CNTF enhances the survival of both photoreceptors and retinal ganglion cells, the two major cell classes damaged in retinal degenerative diseases, a CNTF-based therapy has been developed and tested in several clinical trials. To date, outcomes of the trials for retinitis pigmentosa, geographic atrophy, and achromatopsia have not demonstrated sufficient therapeutic benefits for human patients [2][3][4][5] . However, CNTF therapy for macular telangiectasia type 2 has shown efficacy 6,7 , and a clinical trial for the second largest blinding disease glaucoma 8 is ongoing (https://clinicaltrials.gov/ct2/show/NCT01408472).
To shed light on the molecular and cellular mechanisms of CNTF-mediated neuroprotection, we have delivered the same secreted recombinant CNTF used in human trials to a mouse model expressing the Rds/peripherin2 (Prph2) transgene with the P216L mutation 9 . Pheripherin2 is a structural protein of the photoreceptor outer segment disc, and over 80 PRPH2 pathological mutations have been linked to RD, with clinical presentations ranging from retinitis pigmentosa to various forms of macular degeneration 10 . Transgenic mice expressing the P216L mutant peripherin2 protein exhibit progressive retinal degeneration with rod cell death preceding cone photoreceptor loss, mimicking the dominant disease feature observed in retinitis pigmentosa patients carrying this missense mutation 9 . In the mammalian retina, CNTF activates both the Jak-STAT and MEK-ERK signaling pathways 11 through binding to a tripartite receptor complex consisting of gp130, LIFRβ, and CNTFRα, with

Short-term CNTF treatment-induced transcriptome changes.
To determine effects of the RD mutation-induced perturbation, we performed RNA-seq analyses and compared retinal-transcriptomes of wild type and Rds/Prph2(P216L) 9 (hereafter referred to as Rds) mice at postnatal day 25 (P25) (Supplementary Table S1). To examine the acute impact of exogenous CNTF on retinal gene expression, we also analyzed Rds retinas after shortterm CNTF treatments for 3 hours or 24 hours with parallel PBS-injected Rds retinas as controls (Supplementary  Table S1). RNA-seq data analyses revealed differentially expressed genes between matched sample pairs ( Fig. 1). At P25, when the onset of retinal degeneration had begun in the Rds retina, the mutant retina exhibited significant differences in mRNA expression from wild type littermates for 1322 genes, with 651 genes downregulated and 671 genes upregulated (Fig. 1a). After 3-hour treatment of the Rds mutant, the CNTF-induced response was readily detected in transcript levels of 588 genes compared to vehicle-injected mutant retinas as controls (Fig. 1b). After 24-hour CNTF treatment, 585 genes in total showed significantly altered expression compared to vehicle-injected Rds mutant controls (Fig. 1c). Furthermore, we observed overlapping genes among the significantly altered transcripts in the three sample pairs (Fig. 1d). Among altered transcripts between the Rds and wild type retinas, we found 325 genes also corresponded to short-term CNTF treatment-induced changes, suggesting that the Rds mutation and short-term CNTF treatments shared cellular processes and responses. Interestingly, the gene sets between the 3-hour and 24-hour CNTF treatments were not identical, with an overlapping subset of only 144 genes, reflecting the dynamic retinal transcriptome response to exogenous CNTF signals.
We next compared the top 30 gene signatures for the three matched sample sets of the short-term CNTF treatments ( Fig. 2; Supplementary Table S2). At the onset of degeneration, the P25 Rds mutant already exhibited significant changes in gene expression compared to the wild type retina. As controls for both short-term treatments, the PBS-injected Rds retinas largely retained the transcriptome landscape of the non-injected Rds mutant (Fig. 2a). However, the 3-hour and 24-hour CNTF treatments both resulted in distinct transcriptome alterations (Fig. 2a). We observed several cytokines and growth factors with altered expression during an initial 3-hour CNTF-induced transcription wave, including IL1a, IL6, Cxcl2, Ccl3, Ccl4, Ccl7, and Fgf2. However, this early wave of induced transcripts was diminished within 24 hours and replaced by a different group of elevated transcripts, which included multiple components of the complement system, such as C1qa, C4a and C4b. The two short-term samples also shared common transcripts, including Irgm2, Igtp, D8Ertd82e, Bcl3, and Serpina3f, which were further elevated compared to their corresponding controls. Most of these common genes participate in immune and complement pathways [18][19][20] .
We used MsigDB, a component of Gene Set Enrichment Analysis (GSEA), to further investigate enrichment of the detected gene signatures during short-term CNTF perturbations ( Fig. 2b-d). The analysis showed that genes related to the immune system, cell-matrix interactions, and developmental processes were already overrepresented in Rds versus wild type retinas (Fig. 2b). After short-term CNTF exposure, transcripts associated with immune pathways (e.g. cytokine and interferon signaling), neuronal processes (transmission across chemical synapses and neurotransmitter receptor binding, etc.), growth factor signaling, GPCR signaling, and small molecule transporter systems became the dominant gene signatures (Fig. 2c,d). We also explored the directionality (up or down regulation) of gene expression at the pathway level. The up-regulated gene signatures in short-term CNTF treated Rds retinas were mostly associated with immune-related extracellular growth factor signaling pathways, whereas the down-regulated genes were frequently involved in GPCR signaling and neuronal system processes (Supplementary Table S3).
Rapid impact on cytokine signaling and photoreceptor transcription. We next examined the expression of cytokine signal transduction components and photoreceptor-specific genes (Fig. 3). CNTF rapidly induced expression of the Stat family of genes, especially Stat3 and Stat1 (Fig. 3a). In addition, CNTF augmented cytokine receptor Il6st/gp130 expression within 3 hours (Fig. 3b). Within a 24-hour period, CNTF transiently stimulated Tnfa expression, while substantially elevated expression of Edn2, a factor induced in a variety of retinal degeneration conditions 21 (Fig. 3c). The RNA-seq data also confirmed that Lif was upregulated in the Rds mutant compared to the wild type retina (Fig. 3d). Furthermore, transient spikes of Lif and Clcf1, two members of the CNTF subfamily of cytokines, occurred after a 3-hour CNTF stimulation, but quickly declined to non-treated Rds mutant levels within 24 hours ( Fig. 3d; Supplementary Table S2) 13 .

Influence on transcriptome by long-term CNTF treatment.
To study the long-term effects of CNTF treatment on retinal transcriptomes, we expressed the recombinant human CNTF in vivo using a lentiviral vector LV-CNTF, which expresses the same secreted form of recombinant CNTF used in clinical trials 13 . Since subretinal delivery of LV-CNTF at P25 results in substantial rescue of photoreceptors in the Rds mutant by P35 13 , we performed RNA-seq profiling of wild type, and Rds mutant retinas transduced with either LV-CNTF or a GFP-expressing control virus LV-IG from P25 to P35 (Supplementary Table S1). Retinal transcriptome analyses www.nature.com/scientificreports www.nature.com/scientificreports/ for differential expression detected 238 genes between wild type and Rds mutant retinas infected with LV-IG, 279 genes between wild type and Rds mutant retinas infected with LV-CNTF, and 324 genes between LV-IG and LV-CNTF virus infected Rds retinas ( Fig. 4; Supplementary Table S5).
Analyses of the top gene signatures for each pair-wise comparison revealed shared gene expression trends between LV-CNTF and control virus transduced Rds retinas, as well as distinct gene signatures associated with long-term CNTF treatment (Fig. 5a). Pathway enrichment analyses showed that when compared to wild type retinas, the P35 Rds mutant transduced with LV-IG displayed alterations in gene signatures related to Rap1 and PI3k-Akt signaling, synaptic vesicle cycling, axonal guidance, and fatty acid elongation (Fig. 5b). Predominant differential gene signatures between P35 Rds retinas treated with LV-CNTF and wild type retinas were related to the adaptive and innate immune systems (Fig. 5c). For comparison between LV-CNTF and LV-IG transduced Rds retinas, long -term CNTF treatment retained the top gene signatures related to the immune system as observed with the short-term CNTF exposure (Fig. 5d). In addition, long-term CNTF treatment resulted in transcript enrichment for genes associated with cellular homeostasis, RNA metabolism, insulin signaling, and transcription/translation processes, whereas genes involved in neuronal processes, such as chemical synapse transmission and neuroactive ligand receptor interactions, were down-regulated ( Fig. 5d; Supplementary Table S6).
Commonality of retinal transcriptome changes among degeneration models. In order to search for common transcriptome features for retinal degeneration, we explored existing databases relevant www.nature.com/scientificreports www.nature.com/scientificreports/ to our RNA-seq analyses (Supplementary Table S7). First, we retrieved retina phenotypic genes from the MGI database and found expression overlaps with both our short-and long-term CNTF treatment-induced gene signatures (Fig. 6a). In addition, we detected signatures that overlapped between CNTF-dependent genes and retina regulatory genes 27 (Fig. 6a). Second, we compared the transcriptomes of the Rds mouse model used in this study with RNA-seq data from two other retinal degeneration models, the BC027072 knockout mouse 28 and the Rd10 mouse 29 (Supplementary Table S8). The top gene signatures of the Rds mutant at P25 and P35 showed significant overlap with these two other disease models 28,29 (Supplementary Fig. S1). For instance, 497 out of www.nature.com/scientificreports www.nature.com/scientificreports/ 1079 (46.1%) signature genes from the Rd10 mouse overlapped with Rds mutant signature genes that deviated from wild type (Supplementary Fig. S1). The common enrichments included genes in immune pathways and neurotransmitter-related pathways, suggesting that Rd10 and Rds mutants shared perturbations in similar biological processes during photoreceptor degeneration.
In order to gain insight into potential changes at the protein level in Rds mutant mice under the influence of CNTF, we analyzed and cross-referenced existing proteomic data from the Rd10 model during degeneration 30 . Many gene transcripts significantly impacted by both short-or long-term CNTF treatments in the Rds retina were also detected in the Rd10 retinal proteome (2600 proteins in the pool) at pre-, peak-, and post-degeneration time points, with 69 gene signatures matching Rd10 protein signatures that deviated from wild type at any time point (Supplementary Fig. S2) 30 . In particular, we observed the same declining trend of photoreceptor genes and proteins, including Crx, Rho, Gnat1, Pde6b, Pde6g, beginning at the peak of degeneration, and becoming more severe at the post-degeneration stage. Conversely, a large set of gene transcripts induced in CNTF-treated Rds retinas matched the increased protein levels observed in the degenerated Rd10 retina, including the chemokine signaling pathway genes STAT1, STAT3, ROCK2, GRK1, and GFAP.

Impacts of CNTF treatment on different retinal cell types.
We have shown previously that after one-hour exposure to exogenous CNTF, the Jak-STAT signaling pathway is predominantly activated in Müller glial cells in the mouse retina 13 . However, by 24 hours post CNTF exposure, the signaling events are no longer restricted to Müller glia, but instead have propagated throughout the entire retina. We thus explored whether CNTF-induced transcriptome alterations affected different retinal cell types. We cross-checked our bulk RNA-seq data with the single cell RNA-seq data across ~28,000 mouse retinal cells 31 . This analysis revealed that several different cell types are affected by exposure to exogenous CNTF (Fig. 6b), including Müller glia, rod and cone www.nature.com/scientificreports www.nature.com/scientificreports/ photoreceptors, bipolar cells, and amacrine cells, thus suggesting that diverse cell types and molecular mechanisms contribute to the effects of CNTF on the retinal transcriptome.

Discussion
Phototransduction and the visual cycle are important functions in sensing and converting light signal into biochemical and electrical signals in the retina. Perturbations or mutations of the molecules responsible for these visual processes cause several types of RD. There is great genetic and allelic heterogeneity associated with the various retinal dystrophies 32 . Animal models, especially mice carrying gene mutations, have been used as powerful tools to investigate the etiology of human RD [33][34][35] . RNA-Seq is an extremely high-throughput approach compared with previously available analytic methods, providing the foundation for novel genetic discovery and identification of potential therapeutic targets in the treatment of RD 36 . To understand the molecular mechanism underlying CNTF-mediated neuroprotection, we performed RNA-seq analyses using a well-established preclinical model of retinitis pigmentosa. Our study generated RNA-seq transcriptomes from mouse retinas under different physiological conditions, including wild type, the Rds/Prph2(P216L) mutant, and the Rds mutant exposed to either short-or long-term CNTF treatments. Our results revealed deviations between the Rds and the wild type transcriptomes and commonalities shared among different mouse RD models. In addition, our data captured the retinal transcriptome landscape remodeling that occurs in response to acute and prolonged CNTF treatments in the degenerating retinas.
Through RNA-seq analyses of short-term CNTF treatment, we captured a dynamic transcriptome landscape. Consistent with the progression of photoreceptor degeneration in retinitis pigmentosa, the RNA-seq data showed that at an early stage of degeneration in the Rds retina (P25), the expression levels of rod-specific genes Crx, Nrl, Rho, Gnat1, Pde6a, Pde6e, and Pde6g had already decreased, while the expression levels for cone-specific genes had not been perturbed. After a 24-hour CNTF treatment, not only were the rod-specific transcript levels further decreased, but the mRNAs for cone-specific Opn1sw, Opn1mw, and Gnat2 were also markedly reduced. The rapid and strong CNTF impact observed on key photoreceptor transcription factors and phototransduction apparatuses is likely to cause suppression of visual function in the Rds mutant as observed previously [15][16][17] . In addition, others have reported adverse effects of CNTF on the visual cycle 37 . A short 3-hour CNTF treatment was sufficient to cause a rapid upregulation of transcripts for key cytokine signaling components Stat1, Stat3 and Il6/gp130, suggesting a positive regulation upon CNTF signaling. Consistent with our previous study 13 , a transient burst of the CNTF-family ligands Lif and Clct1 was detected, which returned to the control Rds levels by 24 hours. In contrast, our showed that CNTF induces an increase of Müller glial expressed Gfap and Sox9, indicating that CNTF triggers Müller cell activation. Within 24 hours, CNTF significantly elevated transcripts of Edn2 over the levels detected in none-treated Rds mutant. Edn2 is known to play an important role in photoreceptor-to-Müller glia signaling in damaged or degenerating retinas, as Edn receptor Ednrb is expressed by Müller glia 21 . It has been shown that Edn2 overexpression affect retinal vasculature by inhibiting endothelial tip cells, possibly through Müller released factors 38 . Our RNA-seq data is consistent with a central role of Müller glia in mediating CNTF effects, possibly by releasing multiple growth factors and/or cytokines. It remains to be determined whether CNTF signaling also impacts retinal vessels that can affect nutrients and oxygen supply to the neural retina.
Our gene signature pathway analyses demonstrate a prominent participation of the immune response. Compared to wild type retinas, the Rds mutant retina at the onset of degeneration already shows increased expression of genes involved in the innate immune system, interferon signaling, and the complement pathway. Both short-and long-term CNTF treatments further elevated the expression of genes involved in innate and adaptive immune systems, interferon, NF-kB, Jak-STAT, GPCR and MAPK signaling. Recent studies have shown a rapid and persistent activation of microglia in degenerating retinas, accompanied by significantly enhanced chemokine and cytokine gene upregulations [39][40][41][42] . Part of our RNA-seq data showing the uptick of immune system www.nature.com/scientificreports www.nature.com/scientificreports/ and chemokine gene expression may reflect microglia activities in the retina. However, the many changes we observed to the retinal transcriptome may also reflect CNTF-provoked responses of retinal neurons and Muller glia. By cross-validating our bulk RNA-seq data with a single-cell retinal RNA-seq dataset, we identified several retinal cell types that responded to CNTF treatment in the Rds retina, including rod, cone, bipolar, amacrine, and Müller glia. This result indicates that CNTF broadly impacts different retinal cell types under degeneration conditions, either directly or indirectly. It is also known that different retinal cell types express complement pathway components 43 . Our results captured changes of various complement genes, suggesting that CNTF may regulate this important system and influence their functions in the adult retina.
The mechanisms underlying long-term CNTF-mediated neuroprotection in RD are not well-understood. Interestingly, after long-term CNTF treatment, we detected altered gene signatures in cell homeostasis, ribosomal and RNA metabolism, endocytosis, ER-phagosome pathways, and transcription and translation processes. These gene signatures potentially point to cellular physiological processes influenced by CNTF to enhance neuronal viability. RNA-seq analyses may also capture novel retinal regulatory genes 27 . A recent study of retinal transcriptomes from ~500 age-related macular degeneration (AMD) cases identified ~50 AMD genes 44 . Among these, 12 genes that are involved in complement cascade and extracellular matrix organization, matching the CNTF-treated Rds gene signatures (Mmp9, cd63, Dxo, Cfi, Cfh, Col8a1, Nlrc5, Rdh5, Slc16a8, ApoE, Tsc22d4, and Vtn). It is intriguing that the gene signatures from mouse models can be directly associated with human RD based on genetic studies.
In summary, this study demonstrates that exogenous CNTF alters retinal transcription landscapes in a preclinical animal model of RD. Transcriptome analyses reveal dynamic and strong impacts on photoreceptors and other retinal cell types, as well as the activation of immune systems and multiple signaling pathways. This study provides a molecular simulation at the transcription level for likely responses in diseased human retinas undergoing CNTF treatment, thus shedding light towards developing new targeted therapies for blinding diseases.

Materials and Methods
Animals and eye injections. The Rds/Prph2(P216L) transgenic mice 9 were maintained as hemizygotes on CD1 or C57BL/6J background. Genotyping for Rds/Prph2(P216L) were performed using tail genomic DNAs and PCR primers 5′CCT GGA GTT GCG CTG T, and 5′GTC TTT TTC ATG AAG CAC C. For short-term treatments, Rds/Prph2(P216L) transgenic mice with C57BL/6 background were injected intravitreally with 1 μg of recombinant rat CNTF (PeproTech) or PBS in 1 μl volume at postnatal day 25 (P25), and harvested at 3-hour or 24-hour post injections. For long-term treatment, Rds/Prph2(P216L) mice in the CD1 background were injected subretinally with lentivirus LV-CNTF or the control virus LV-IG 13 at P25, and harvested at P35. All animal procedures followed National Institutes of Health guidelines and were approved by the Animal Research Committee at University of California Los Angeles.
RNA extraction, amplification, and sequencing. The RNeasy RNA extraction kit was used to isolate total RNA from dissected neural retinas without the retinal pigment epithelium and other ocular tissues. On-column DNaseI digestion was performed and RNA was collected in RNase-free water, following manufacturer's instructions (Qiagen, Hilden, Germany). The RNA quality was determined on an Agilent 2100 Bioanalyzer. All samples passed the quality filters. For short-term treatments, RNA was treated with oligo-dT beads to enrich for mRNA out of total RNA, and the libraries were preparing using Illumina TruSeq RNA Sample Preparation Kit and loaded onto a HiSeq2000 Sequencer for single-end sequencing with read lengths of 50 bases. For long-term treatment, sequencing libraries were prepared using the Illumina TruSeq Stranded Total RNA with Ribo-Zero Gold Library Prep kit. Libraries were loaded onto the HiSeq4000 Sequencer for paired-end sequencing with read length of 69 bp. The datasets are summarized in Supplementary Table S1. RNA sequencing data analyses. Quality of individual sequences were evaluated using the FastQC software (www.bioinformatics.babraham.ac.uk/projects/fastqc/) after adapter trimming with the cutadapt software (cutadapt.readthedocs.io/en/stable/). The reads were filtered to exclude those with more than 5% unknown bases, reads which contained more than 20% bases with quality score below 15, and reads with only adapters. After filtering, the median size was 5.65 Gb per library (range 3.66-7.26 Gb). Sequenced reads were then aligned to the Mouse reference mm10 from UCSC (genome-euro.ucsc.edu) using Hisat2 (v2.0.4) 45 . The expression levels were normalized by calculating the fragments per kilobase million reads (FPKM) values. The differential expression of transcripts was calculated using DESeq2 46 , with statistical significance p < 0.01. The pathway enrichment analyses were performed using GSEA 47 and databases from the Molecular Signatures Database (MSigDB v7.0, http://www. broadinstitute.org/gsea/msigdb/).
Validations of gene signatures using bulk RnA-seq, single-cell sequencing, and proteome data sets. To compare the gene signatures with other studies of the mouse retina, we retrieved results from previously published studies, including (1) retina RNA-seq data of two RD models 28,29 , (2) single-cell RNA-seq data of the mouse retina 31 , and (3) proteome data from the Rd10 model 30 (Supplementary Table S7). First, we explored diverse RD mouse models with RNA-seq transcriptomes available in the GEO database (www.ncbi.nlm.nih.gov/ geo/) and retrieved studies with ≥3 biological replicates for both wild type and mutant groups (GEO accessions: GSE56473, GSE63810) 28,29 . These RNA-seq data sets were reprocessed using the Hisat2-DESeq2 pipeline described above. Second, we retrieved single-cell sequencing data of mouse retina (GEO accession: GSE81905) 31 to evaluate which cell types are enriched in the gene signatures we detected. Third, we cross-validated the transcriptome gene signatures with Rd10 proteome dataset at pre-, peak-, and post-degeneration time points (ProteomeXchange accession: PXD002584) 30 .