RNA-sequencing indicates that the neural retina favors the NAD-salvage pathway for NAD synthesis
To determine which NAD-synthesizing enzymes are present in the retina we queried publicly available RNA-sequencing data to determine which genes are transcribed. We first examined whole retina bulk RNA-sequencing data from The Genotype-Tissue Expression (GTEx) Project through The Human Protein Atlas project portal, allowing for an assessment of transcripts in human retina across a sample of 105 individuals. NAD-salvage pathway transcripts (NAMPT, NMNAT1, NMNAT2, NMNAT3) were expressed in whole retina from all individuals as was NMRK1, but not NMRK2 (detected in only 32 individuals, of which all had negligible expression of < 0.5 normalized transcript per million (nTPM)). NAMPT and NMNAT3 were highly variable between individuals. The Preiss-Handler pathway transcript NAPRT was lowly expressed and only detected in 21 individuals (< 20%) but NADSYN1 was well expressed (Fig. 1B).
Next, we determined which cells express these transcripts by examining publicly available single cell sequencing data from human retina. We examined two independent datasets: a single cell RNA-sequencing dataset (scRNAseq) from 4 individual postmortem normal retina (~ 15,000 cells) and single nucleus RNA-sequencing dataset (NucSeq) from 4 individual postmortem normal retina (~ 100,000 nuclei in total). These allowed for identification of expression in retinal neurons (cones, rods, horizontal cells (HCs), bipolar cells (BPs), amacrine cells (ACs), retinal ganglion cells (RGCs)), and non-neuronal retinal cells (myeloid/microglia and Müller glia, and additionally astrocyte, vascular cells, and retinal pigment epithelial cells (RPE cells) in the NucSeq; Fig. 2). NAPRT was lowly expressed in only a small percentage of neurons in the scRNAseq and NucSeq (Fig. 2A). In contrast, NAMPT expression was greater in all retinal neurons (Fig. 2A) indicating a preference for the NAD-salvage pathway over the Preiss-Handler pathway. Likewise, NADSYN1, the terminal enzyme of the Preiss-Handler pathway, was expressed in very few retinal neurons compared to NMNAT1-3 (Fig. 2A). NMRK1 expression was detected in very few neurons, and NMRK2 was almost entirely absent, indicating a preference for nicotinamide (NAM) as a substrate over nicotinamide riboside (NR) in retinal neurons (Fig. 2A). Of these, RGCs had both the highest average expression and highest percentage of cells expressing NAD-salvage pathway transcripts (Fig. 2A). NAMPT and NMNAT1 appear to be particularly important in RGCs over other retinal neurons, while NMNAT2 expression was exclusive to RGCs in the scRNAseq, with NucSeq demonstrating that other retinal neurons do express NMNAT2, but to a lesser extent (Fig. 2B). Examining the distribution of expression (Fig. 2C, ridgeplots) revealed NAMPT and NMNAT1 expression to be normally distributed, whereas NMNAT2 had a greater variance, perhaps indicating RGC-subtype specific variation in NMNAT2. NMNAT3 expression was greatest in rods and cones, followed by RGCs (Fig. 2A–C), given that NMNAT3 mRNA has a mitochondrial targeting sequence this would appear to fit with the high density of mitochondria in these neurons. However, NMNAT3 is known to be translationally repressed due to an upstream open reading frame in the mRNA 5′UTR region and mature protein has only been identified in cells following over-expression through plasmid transfection [17]. Although rods, cones, and RGCs express NMNAT3 it is, therefore, unlikely to have a functional role at a protein level in these cells.
In the non-neuronal cells of the retina, all transcripts were expressed in a lower percentage of cells than in neuronal cells, with the exception of NAMPT and NMNAT3 (Fig. 2A–C), perhaps reflecting the lower metabolic demands of these cell types relative to neurons. NAMPT was expressed by a higher percentage of microglia/myeloid cells, astrocytes, Müller glia, vascular cells, and RPEs than in neuronal cell types; its average expression was greatest in microglia/myeloid cells (Fig. 2A, B). Expression of NAPRT, NMRK1, and NADSYN1 was generally greater than in retinal neurons (except RGCs) demonstrating that non-neuronal cells in the retina may have a greater flexibility in NAD production by utilizing more substrates and pathways (Fig. 2A, B). Supporting this, when we expanded our examination to other cell types of the eye in the scRNAseq (e.g. corneal epithelial cells, ciliary body cells) we identified that NMRK1 and NADSYN1 average expression and percentage of expressing cells is far greater in the anterior chamber of the eye (Additional file 2: Fig. S1). This suggests that while the retina favors the NAD-salvage pathway, the anterior chamber favors the Preiss-Handler pathway and nicotinamide riboside as an alternative substrate to the NAD-salvage pathway.
Antibody labelling identifies the presence of NAD salvage pathway machinery in normal retina and optic nerve
Identification of transcript expression does not determine that a mature protein is expressed since cells employ multiple mechanisms of translational regulation. No protein data is available from the Human Protein Atlas project for any of the NAD-producing enzymes. We therefore utilized a unique resource of enucleated human eyes from the St. Erik Eye Hospital Ophthalmic Pathology archive. We previously identified and characterized 6 glaucoma and 12 control (uveal melanoma, since healthy eyes are not enucleated) eyes from this archive [13]. Importantly, since these eyes are live enucleations and fixed immediately, there are no post-mortem degenerative confounders. We used immunohistochemistry to determine localized protein expression across the retina. We focused on the NAD-salvage pathway since this was the predominant pathway identified through RNA-sequencing. NAMPT was distributed across all layers of the central and mid-peripheral retina, but was greatest in the nuclear layers, particularly the INL (n = 11 control eyes; Fig. 3A). NMNAT1 was clearly confined to the nuclear layers and overlapped with hematoxylin, consistent with its known nuclear localization (n = 11 control eyes; Fig. 3B). Conversely, NMNAT2 labelling was largely uniform across nuclear and plexiform layers and had the highest relative intensity in the RNFL compared to the rest of the retina of all the antibodies tested. NMNAT2 labelling was greatest in the GCL, this was more pronounced in the central, compared to mid-peripheral, retina where RGC density is higher (n = 11 control eyes; Fig. 3C).
NAMPT labelling was present in the both the axon and glial compartments of the optic nerve but was more intense in the latter; the labelling was highly uniform along the optic nerve length (n = 11 control eyes; Fig. 3D). NMNAT1 (n = 10 control eyes) and NMNAT2 (n = 8 control eyes) labelling demonstrated contrasted localization with NMNAT1 located only in nuclei (particularly high in glia compartments; Fig. 3E) and NMNAT2 only in axons (no labeling within or around nuclei, as can be seen in Fig. 3F inset). While NMNAT1 labelling was highly uniform along the optic nerve (Fig. 3E), NMNAT2 was significantly greater in the initial 500 µm (representing the pre-lamina, lamina cribrosa and initial pos-laminar region; Fig. 3F). This may reflect the increased density of axons relative to glia and connective tissue here, as well as the optic nerve head representing a particularly metabolically active portion of the optic nerve where axons first become myelinated.
NAD salvage pathway machinery labelling is reduced in glaucoma
To determine if the expression patterns of these NAD salvage pathway machinery are altered in glaucoma, we compared the control eyes with 7 glaucoma eyes. Since the glaucoma tissue is from late-stage disease, we accounted for remodeling of the inner retinal layers by delineating the whole ganglion cell complex (GCC; combined RNFL, GCL, and IPL) as is common for in vivo OCT imaging studies. NAMPT labelling in the central retina was significantly reduced in the GCC in glaucomatous samples relative to controls and was not significantly altered in the INL or outer retina (n = 11 control eyes, 7 glaucoma eyes; Fig. 4A). NMNAT1 was significantly reduced in the GCC and INL in both central and mid-peripheral retina, with no detectable difference in the outer retina (n = 11 control eyes, 7 glaucoma eyes; Fig. 4B). NMNAT2 labelling was significantly reduced only in the central retina GCC in glaucomatous samples (n = 11 control eyes, 7 glaucoma eyes; Fig. 4C). These changes likely reflect the loss of RGC axons, cell somas, and dendrites, and potential subsequent structural and metabolic remodeling of the INL. In the ONH NAMPT was significantly reduced in the first 500 µm and was otherwise unchanged across the remaining nerve (n = 11 control eyes, 5 glaucoma eyes; Fig. 4D). This pattern was also evident for NMNAT1 (n = 10 control eyes, 5 glaucoma eyes) and NMNAT2 (n = 8 control eyes, 5 glaucoma eyes; Fig. 4E, F). These changes are predominantly a reflection of the loss of neural content through cupping and excavation of the ONH in this late stage of disease.
Taken together, these findings demonstrate that the inner retina and proximal optic nerve/optic nerve head is highly enriched with the machinery to directly utilize nicotinamide through the salvage pathway. However, while the ability to utilize nicotinamide remains in glaucoma, the capacity to do so may be lower during disease.