Do Different Neurons Age Differently? Direct Genome-Wide Analysis of Aging in Single Identified Cholinergic Neurons

Aplysia californica is a powerful experimental system to study the entire scope of genomic and epigenomic regulation at the resolution of single functionally characterized neurons and is an emerging model in the neurobiology of aging. First, we have identified and cloned a number of evolutionarily conserved genes that are age-related, including components of apoptosis and chromatin remodeling. Second, we performed gene expression profiling of different identified cholinergic neurons between young and aged animals. Our initial analysis indicates that two cholinergic neurons (R2 and LPl1) revealed highly differential genome-wide changes following aging suggesting that on the molecular scale different neurons indeed age differently. Each of the neurons tested has a unique subset of genes differentially expressed in older animals, and the majority of differently expressed genes (including those related to apoptosis and Alzheimer's disease) are found in aging neurons of one but not another type. The performed analysis allows us to implicate (i) cell specific changes in histones, (ii) DNA methylation and (iii) regional relocation of RNAs as key processes underlying age-related changes in neuronal functions and synaptic plasticity. These mechanisms can fine-tune the dynamics of long-term chromatin remodeling, or control weakening and the loss of synaptic connections in aging. At the same time our genomic tests revealed evolutionarily conserved gene clusters associated with aging (e.g., apoptosis-, telomere- and redox-dependent processes, insulin and estrogen signaling and water channels).

age-related mechanisms also take into account a high level of conservation among many epigenetic processes known to be lost in nematodes and fl ies which have extremely short lifecycles and particularly derived genomes (Figure 1).
Most importantly, Aplysia posses the largest nerve cells (ranging from 30 to 1,100 µm, see Figure 2) in the entire animal kingdom with only eggs being larger (Kandel, 1976) and truly gigantic growth cones (up to 630 µm, Lovell and Moroz, 2006). These cells can be uniquely identifi ed and mapped in terms of their well-defi ned interactions with other neurons, and have a limited number of recognized synaptic connections forming relatively simpler neural circuits underlying several stereotypic and learned behaviors (Kandel, 1976(Kandel, , 1979. As indicated in numerous studies dealing with memory mechanisms, synaptogenesis (Kandel, 2001;Hawkins et al., 2006) and initial genomic characterization  Aplysia has many parallels and similarities with learning and aging in mammalian models and humans including the presence of NMDA receptors (Moroz et al., 1993;Ha et al., 2006) and nitric oxide signaling pathways (Katzoff et al., 2002(Katzoff et al., , 2006Bodnarova et al., 2005;Moroz, 2006;Antonov et al., 2007).
The phenomenology of aging dynamics in Aplysia has been studied in suffi cient detail both in nature and under controlled aquaculture environments (Stommes et al., 2005;Gerdes and Fieber, 2006). Aplysia has a moderately short lifespan with a well

INTRODUCTION: APLYSIA AS AN EMERGING MODEL IN THE BIOLOGY OF AGING
The complexity of events associated with age-related changes in nervous systems cannot be overestimated. The problem is further complicated by the enormous diversity of neurons in the central nervous system (CNS) and even in synapses of one neuron within a neural circuit. Large scale single-neuron analysis is not only challenging but mostly impractical for any model currently used in the neurobiological analysis of aging or age-related memory loss. We simply do not know: Do all neurons and synapses age differently or are some neurons (or synapses) more resistant to aging than others? What is happening in any given neuron while it undergoes "normal" aging? What are the genomic changes that make aging apparently irreversible? What would be the balance between neuron-specifi c vs. global genome-wide changes in aging? In this paper we begin to address these questions by developing a new model to study the entire scope of genomic and epigenomic regulation in aging at the resolution of single functionally characterized cells and even cell compartments.
Biology has often advanced by making use of relatively simple model systems. We selected Aplysia californica as the foremost experimental organism for the goals of the program. The sea slug, Aplysia is a representative of the largest superclade of bilaterian animals (Lophotrochozoa) which include annelids, fl at worms and about ten less known phyla. The proposed long-term regulatory as their neural substrates, are age-sensitive. In contrast, a third, gill respiratory pumping movements and its neurons, is age-invariant Peretz, 1988).
It is still unclear why some neurons and synapses are age resistant -perhaps because specifi c functional studies on individual neurons were not linked to molecular analysis of the same cells.
Here we have implemented a single cell expression profi ling approach by focusing upon only two classes of cholinergic motor neurons in Aplysia as the proof-of-concept type of study. The unique features of these symmetrical neurons (Figure 2) allow one to perform and validate cell by cell comparisons. Our initial analysis indicates that different cholinergic neurons revealed highly differential genome-wide changes following aging, suggesting that on the molecular scale different neurons indeed age differently. In addition, we have identifi ed a number of evolutionarily conserved and group-specifi c genes that are age-related, including components of apoptosis and chromatin remodeling.

ANIMALS AND SINGLE CELL ISOLATION
Specimens of Aplysia californica were obtained from the Aplysia Facility at the University of Miami and maintained for up to 1 week in marine aquaria with running sea water at 16-18°C and fed twice per week with an unrestricted diet of Gracilaria. All dissection and single cell RNA isolation protocols were as described elsewhere . Briefl y, the animals were anesthetized by injection of 370 mM MgCl 2 . The CNS was immediately dissected; connective tissues from individual ganglia were removed after 45 min treatment with trypsin (1%), washed twice in cold artifi cial sea water (460 mM NaCl; 10 mM KCl; 55 mM MgCl 2 ; 11 mM CaCl 2 ; 10 mM HEPES, pH 7.6) and then quickly transferred to 100% ethanol. Since the selected identifi ed neurons (LPl1 and R2) are among the largest known neurons and are easily identifi ed at the ganglionic surface, these cells can be unbiasedly recognized and physically isolated using glass microelectrodes within 1-3 min as shown in Figure 2. Immediately after the isolation step each of the identifi ed neurons was transferred to 100 µl of lysis buffer (Cat # 1291, Ambion) and RNA isolation was performed using an RNA isolation kit (RNAqueous-Micro; Cat # 1291, Ambion). The quality control and quantifi cation of freshly isolated CNS was performed using a Bioanalyzer and by spectrophotometer (Gene Spec III, MiraiBio Inc., Japan) and RNA was stored at −80°C. Since one cell can produce on average high nanogram to low microgram quantities of total RNA (see Table S1 in Supplementary Material or GEO 18783), this amount is suffi cient for both microarray and , mechanosensory pleural neurons (SN) and motor pedal neurons. LPl1 is located in the left pleural ganglion (red lines represent major axonal branches) and R2 in the abdominal ganglion (blue lines represent major axonal branches). Both the positions of these neurons and their axonal pathways are highly conserved across individuals, and these cells can be reliably identifi ed in closely related species of the genus Aplysia (modifi ed from Hughes and Tauc, 1961;Kandel, 1976) (see details in the text). (B) A photograph of the freshly dissected right abdominal semi-ganglion with the positions of R2 and R14 neurons marked (connective tissues from the ganglionic surface were removed and the natural coloration of cell somata were preserved). This R2 cell is the largest neuron ever photographed reaching 1.1 mm in diameter. When this cell was isolated (insert) and fi xed in 100% ethanol it lost its pigmentation. 1.965 µg of total RNA was obtained from this neuron. These cells can also provide from 150 to 250 ng of DNA depending upon cell size.
control quantitative real-time PCR (QRT-PCR) tests. Labeling and microarray hybridization experiments were performed within 1-3 days after RNA isolation.

MICROARRAY EXPERIMENTS AND DATA ANALYSIS
In the microarray experiments we used two custom 44,000 oligonucleotide arrays which were constructed in collaboration with Agilent Technologies using 60-mer oligonucleotide probes designed from each nonredundant sequence in the Aplysia neuronal EST database as described earlier . The data including materials and methods discussed in this publication have been deposited in NCBIs Gene Expression Omnibus (GEO) and are accessible through the GEO Series accession number GSE18783 with platform accession numbers; GPL3635 record: Aplysia Annotated Array AAA and GPL3636 record: Aplysia Discovery (un-annotated) Array DAA. Additional details about the protocols used including a list of features on each array can be found at the GEO web site.
Three types of array experiments were designed. Two direct comparison array experiments were initially performed. First, we compared gene expression profi ling of the entire abdominal ganglion (approximately 1,000 neurons) from young (∼3-4 months) vs. older (∼7-10 months) animals. The age in commercially reared Aplysia is measured as the time from larvae hatching. That can be detected directly since the age of egg masses and hatching is routinely recorded.
Second, we made direct comparison of single individually isolated R2 cells from young animals (∼3-to 4-month-old, 25-40 g) with R2 isolated from older animals (∼7-10 months, 1000-1500 g). Third, a reference design scheme was implemented ( Figure 3A) in order to compare age-related changes in two different cell types R2 and LPl1, both cholinergic motor neurons (Figure 2A). In this experiment we combined RNA isolated from the CNS (all nine central ganglia, from four middle-age animals: 200-250 g) as a reference platform. Thus, we estimated age related changes in two different neurons to a reference sample. Importantly, here we used the very same material from individual "old" R2 neurons as in the second experiment with direct comparison of R2 young and R2 old (different young R2s were used because of less RNA obtained from these cells).
In all cases three biological replicates were performed. All data are deposited at GEO Series accession number GSE18783. For a list of the animal weight, diameter of cell isolated, and the amount and concentration of RNA extracted see Table S1 in Supplementary Material at GEO GSE18783.
False Discovery Rate (FDR) was performed using signifi cance analysis of microarrays (SAM -a statistical technique for fi nding signifi cant genes in a set of microarray experiments). Analysis for the direct comparison array experiments was performed by signifi cance tests with a 5% FDR and 2-fold change cut-off. All descriptions of the tests and statistical analysis are also deposited at GEO Series accession number GSE18783.
The reference design microarray experiment was analyzed by a standard multivariate analysis of variance (MANOVA) that is a generalized form of univariate analysis of variance (ANOVA). This analysis is used in cases where there are two or more dependent variables as well as identifying whether changes in the independent variable(s) have signifi cant effects on the dependent variables (MANOVA is also used to identify interactions among the dependent variables and among the independent variables). Our MANOVA model has three factors (or effects): Tested Cells/treat (R2, LPl1, CNS), Age (Young, Old, Reference -middle age), and Cells/*Age. Our major focus was on single cell specifi c expression changes as well as their changes as a function of age. The reported column correlates are GENENAME, HITDEF, best hit defi nition, EVALUE evalue for the best hit defi nition, ABUNDANCY the abundance of the target sequence in EST collection Effect, TREAT_TREAT, FOLD CHANGE which is an estimated log ratio between the two samples: log (Old R2/Young R2), base e = 2.718 StdErr the error bar of Estimate. Probt is the probability that the effect is just random noise. A smaller Probt means the effect is signifi cant in the differential expression among different samples. All data were trimmed to ≥2-fold Change and a Probt ≤0.05 or a 95% confi dence level which means <5% false positives. All data and their analysis for each microarray feature is also deposited at GEO Series accession number GSE18783.

CLONING OF TRANSCRIPTS OF INTEREST
Selected transcripts showing differential expression in the array experiments were cloned, then their expression was validated by QRT-PCR or in situ hybridization (Tables 2 and 3). We used previously reported Aplysia EST data  for the initial probe design. These probes from the transcripts of interest were used as the starting points for cloning of all transcripts. Both 5′ and 3′ RACE were performed to obtain the full-length cDNA transcripts which were cloned into pCR 4-TOPO (Cat # K4575-01, Invitrogen) then eight clones were sequenced. Full-length cDNA sequence for AcAPP (GenBank accession number AY535409) was obtained using terminal primers: 5′-CACCAACGTAGTGGAGGCCTG-3′ and 5′-CTCAAAGTATTTGTACGTGGGGT-3′ with amplifi ed cDNA from the CNS and buccal ganglia libraries. Full-length cDNA sequence for AcPSEN1-1 (GenBank accession number AY535407) was obtained using terminal primers: 5′-ATATTGACTGCATGAATCC-3′ and 5′-GAAAGTGACAGTCGTACGCTAT-3′ with amplifi ed cDNA from the CNS library. The full-length transcripts were deposited in GenBank with the accession numbers listed in Tables 1 and 2.

TREE CONSTRUCTION AND SEQUENCE ANALYSIS
The phylogenetic tree was generated using default parameters and 10,000 iterations of the maximum likelihood algorithm implemented in the program TREE-PUZZLE 1 . The initial multiple alignment was done using ClustalX (Thompson et al., 1997;Jeanmougin et al., 1998) with default parameters; all gaps were removed manually in GeneDoc (Nicholas et al., 1997) prior to tree construction. The numbers at branches represent bootstrap values for 10,000 iterations. Branch-length scale bar represents 0.1 amino acid substitutions per site. The graphical output was generated using Treeview (Page, 1996). All protein predictions were determined with Prosite 2 and SMART 3 .

QUANTITATIVE REAL-TIME PCR
Quantitative real-time PCR was used to validate differentially expressed transcripts on the arrays. RNA was extracted from similar single cells of young or old animals as described above using RNAqueous-Micro (Cat # 1291, Ambion). The RNA is DNased as described in the RNAqueous-Micro kit. cDNA synthesis was produced with random hexamers as the primers and SuperScript™ II (Cat # 18080-093, Invitrogen). cDNA products were diluted and aliquoted to 100 ng total RNA input. Sequences of the genes of interest were loaded into Primer Express® software (Applied Biosystems) and primers designed. Amplicon lengths were between 75-125 bp. Two-Step RT-PCR was performed in which the RT and PCR were done separately with the SYBR Green PCR Master Mix (Applied Biosystems, catalog # 4309155) on an ABI 7700 Sequence Detection System (Applied Biosystems). The linear dynamic range of the input RNA (cDNA) from 1 ng to 10 pg was determined. All primer sets were tested for optimal dissociation curves and amplifi cation effi ciencies between 85 and 100% and all primer sets not meeting the standards were redesigned. All runs were normalized to 18s RNA, our endogenous control. The relative standard curve method was employed for analysis and an expression ratio calculated for each sample pair. All data were performed in quadruplicate and statistics analyzed by the relative standard curve method. Aplysia amyloid-like B peptide shares higher identity to the human APP, BAA22264, APLP1, NP_ 005157.1 and APLP2, NP_001633.1, including the putative cleavage site by γsecretase, than C. elegans, AAK68242 or Drosophila, NP_476626.2. Shark, Narke japonica BAA24230.1, zebrafi sh, Danio rerio AAK64495, and rat, Rattus norvegicus AAH62082.1 are included. Tree construction and sequence analysis. The phylogenetic tree was generated using default parameters and 10,000 iterations of the maximum likelihood algorithm implemented in the program TREE-PUZZLE (http://www.tree-puzzle.de). The initial multiple alignment was done using ClustalX (Thompson et al., 1997;Jeanmougin et al., 1998) with default parameters; all gaps were removed manually in GeneDoc (Nicholas et al., 1997) prior to tree construction. The numbers at branches represent bootstrap values for 10,000 iterations. Branch-length scale bars represent 0.1 (APP) and 0.05 (presenilins) amino acid substitutions per site. The graphical output was generated using Treeview (Page, 1996). All protein predictions were determined with Prosite (www.expasy.org/cgi-bin/scanprosite) and SMART (www.smart. embl-heidelberg.de

DIRECT SINGLE CELL 454 PYROSEQUENCING
We used an unbiased method of library construction, applicable for single neurons, for transcriptional profi ling that is both quantitative and qualitative and that preserves the direction of transcripts. We used commercially available kits (Marathon® cDNA Amplifi cation Kit, Cat #634913, Clontech) to ensure consistency and reproducibility. Methods for single neuron manipulation, RNA extraction, and RNA quantifi cation were the same as described above. Due to the limited amount of material for single cell applications total RNA quantifi cation and quality is critical. For cDNA library constructions we used one cholinergic R2 neuron, one L7 motor neuron, three serotonergic MCC neurons and a cluster of approximately 200 sensory neurons. Library construction starts with total RNA reverse transcribed to cDNA with an oligo dT primer, then second strand generated. The double stranded cDNA is fragmented with the restriction enzyme Alu1 (Cat # R0137s, New England Biolabs Inc.) then 454 specifi c adaptors are ligated onto the double-stranded cDNA. The 5′ or 3′ preferential directions of the constructed libraries are determined by timing of the digestion and the ligation of adaptors. Fragmented DNA with ligated adaptors is processed through an emulsion-based clonal amplifi cation (emPCR) then captured onto beads as required for subsequent sequencing steps. DNA-captured beads are placed on a PicoTiterPlate device for pyrosequencing using the 454 FLX platform. The sequencing method is both qualitative and quantitative as well as preserving the direction of transcripts. We validated the quantitative aspect with absolute RT-PCR and spiked in controls; the direction was similarly validated with spiked in controls. In summary, we obtained over 2.3 million sequences or 321 million bases from these four neuronal libraries. These data allow us to quantify a given transcript by obtaining the number of reads specifi c to that transcript and expressing a % relative to the total number of sequences for the neuron of interest ( Figure 5A). In addition the sense and antisense of any specifi c transcript can be determined and reported as % antisense ( Figure 5B).
In situ hybridization for whole mount CNS from Aplysia was performed as described in detail elsewhere (Jezzini and Moroz, 2004;Walters et al., 2004;Jezzini et al., 2005). Briefl y, we used cloned full-length cDNA from Table 2 for the preparation of in situ probes. The antisense probe was generated by digesting plasmid DNA with NotI (Cat # R0189s, New England Biolabs Inc.) then transcribed with T 3 polymerase from the DIG RNA Labeling Kit (Cat # 11 175 025 910, Roche Diagnostics). The control sense probe was produced by the same protocol but used Pme1 (Cat # R0560s, New England Biolabs Inc.) to digest the cDNA and T 7 polymerase for transcription. The DIG-labeled antisense probes were hybridized to whole mount CNS preparations, and the neurons containing the probe-target duplex were localized and visualized with alkaline phosphatase-conjugated anti-DIG antibody fragments (Cat # 11 093 274 910, Roche Diagnostics). Expression patterns in the CNS were investigated in all central ganglia from at least three experimental (antisense probe) and three control (sense probe) CNS preparations. Control in situ hybridization experiments with full-length sense probes revealed no specifi c and selective staining in the CNS under identical conditions and labeling protocols. Images were acquired with a Nikon Digital Sight DS-5M digital camera mounted on an upright Olympus SZX12 microscope.

IDENTIFICATION OF AGE-RELATED GENE HOMOLOGS IN APLYSIA
Aplysia, similarly to many other representatives of the superclade Lophotrochozoa, has a relatively slow rate of evolution of its genes, suggesting preservation of an ancestral subset of genes compared to faster evolving genomes as in nematodes and fl ies (Figure 1). Thus, it is likely that Aplysia neurons possess a diversity of gene homologs known to be associated with aging diseases in humans, including Alzheimer's. We addressed this question by both direct sequencing from specifi c cells and analysis of gene expression patterns using microarrays. As a result, we have identifi ed a suite of human age-related homologs in Aplysia ( Table 1). All of these genes have recognizable and highly conserved motifs with their human counterparts. The reconstructed phylogenetic trees (Figures 3A,C) show that the evolutionary distance (represented by branch length) from Aplysia to humans is shorter than the distance from some other invertebrate models to humans. This indicates that the amino acid replacement rate has been lower in the lineage leading to Aplysia than in the lineages leading to Drosophila and C. elegans for example.
Previous comparative analysis of evolutionarily conserved genes supports this hypothesis: greater similarity was found between human and Aplysia proteins, confi rming a slower rate of protein evolution than in the lineages leading to Drosophila and C. elegans or sea urchin . It was also shown that more than a hundred homologs linked to human neurological diseases are expressed in the CNS of Aplysia  while a fraction of them were lost in fl ies and nematodes.

Aplysia homologs of genes associated with Alzheimer's disease
Two prominent Aplysia homologs of the Alzheimer's disease (AD) related transcripts are amyloid precursor protein (AcAPP) and presenilins or gamma-secretases. In the etiology of AD, cleavage of the APP by β and γ secretases leads to the deposition of an amyloid β peptide (Aβ) as plaques in brain tissue of AD patients (Selkoe, 1991(Selkoe, , 1999Hardy and Higgins, 1992;Hardy and Selkoe, 2002;Walsh et al., 2007). AcAPP cDNA is a 2074-bp transcript that encodes a putative 668 amino acid protein. The Aplysia AcAPP shares 29% overall identity with the human APP while the predicted Aplysia Aβ like peptide has higher (∼50%) identity to the human Aβ peptide (including the putative secretase cleavage site - Figure 3B) compared to Drosophila at 15% or C. elegans at 4%. The predicted AcAPP contains all functional domains such as the A4_EXTRA and β-amyloid precursor protein C-terminus domains. Phylogenetic analysis for the family of APP genes implies that at least two gene duplication events occurred for the APP gene family: the vertebrate APP, APLP2 duplication event and an earlier duplication event for the APLP1 genes ( Figure 3A). The majority of the invertebrate APP gene family forms a separate lineage that is more basal to the others.
The AcAPP was widely distributed across the CNS. In fact, it is one of the most highly expressed transcripts, comparable in abundance to cytoskeleton proteins and selected neuropeptides. The AcAPP is expressed in a majority of neurons throughout the CNS but it is not uniformly distributed, having prominent expression in several groups of motor neurons (e.g., E-and A-/B-cluster cells Figure 4D) such as cholinergic neurons (including R2 and LP1, Figure 4B) as well as in modulatory serotonin containing interneurons (MCC, Figure 4D) known to be involved in feeding arousal (Weiss and Kupfermann, 1976;Kupfermann et al., 1979;Kupfermann and Weiss, 1982).
In mammals, presenilins 1 and 2 (PSEN1 and PSEN2), which are processing enzymes of APP and the ApoE4 protein, are implicated in the etiology of AD (Walsh and Selkoe, 2004;Selkoe and Wolfe, 2007). In Aplysia we indentifi ed two putative isoforms of Presenilin 1-1 and 1-2 (AcPSEN1-1 and 1-2); each contains all 8 transmembrane regions and the peptidase_A22B domain. These predicted proteins AcPSEN1-1 (581aa) and 1-2 (561aa) share high identity with each other and 46% identity with human PSEN1. Two highly conserved aspartic acid residues are necessary for γ-secretase function and are present in both AcPSEN1s at Asp291, Asp466 for AcPSEN1-1 and Asp271, Asp446 for AcPSEN1-2 (Tandon and Fraser, 2002). The presenilin family of proteins appear to be phylogenetically ancient and well conserved across a diverse range of eukaryote species, including slime molds, choanofl agellates (as a sister group for animals), basal metazoans, molluscs, insects, fi sh and vertebrates ( Figure 3C). In addition to the reported data for Aplysia, it is only mammals that have the closely related PSEN2.
In contrast to APP, presenilins are less abundant in the CNS. The in situ hybridization shows very localized expression in the CNS with the highest levels for the AcPSEN transcript in the F-cluster (Rubakhin et al., 1999) -a neurosecretory center involved in feeding arousal and control of animal energetics (Figure 4C).

Neuron-specifi c expression of sense and antisense transcripts
Analysis of the relative abundance of AD related and other cloned transcripts has also revealed enormous cell specifi city in their expression including sensory (SN) and motor (L7) neurons of a simple Aplysia memory forming circuit (Kandel, 1976(Kandel, , 2001Walters et al., 2004). Since the implemented 454 sequencing libraries are directional, we were able to compare presence and relative expression level of natural sense and antisense transcripts within the same identifi ed neurons ( Figure 5).
Remarkably, the distribution of antisense transcripts is also neuron-specifi c. For example, it appears that all tested neurons do express a relatively low fraction of sense Tau transcripts (Tau mRNA) while antisense Tau transcripts are predominant in these cells (reaching 95-98% of the total amount of sense and antisense Tau transcripts). Similar relationships are observed for transcripts encoding cloned insulin degrading enzyme where the fraction of antisense transcripts can be as high as 50-75% of corresponding mRNAs. High abundance of antisense transcripts for ApoE receptor and huntingtin were found in the L7 motoneuron but not other tested cells. In contrast, APP and major vault proteins (a class of RNA binding proteins) mRNAs (sense) had relatively elevated abundances compared to corresponding antisense transcripts ( Figure 5B). The nature and functional role of such cell specifi c antisense transcription is unclear but an inhibitory regulatory role in control of translation has been suggested and tested for other classes of RNAs (Katayama et al., 2005;He et al., 2008;Mercer et al., 2008Mercer et al., , 2009Brosnan and Voinnet, 2009;Mattick, 2009;Ponting et al., 2009;Szulwach et al., 2009;Wilusz et al., 2009).
In summary, the initial survey of potential age-related genes in Aplysia revealed highly differential levels of expression in a majority of tested neurons suggesting that transcriptional changes in the course of normal aging may also be cell-specifi c. To further test this hypothesis, we selected two identifi ed cholinergic motor neurons named R2 and LPl1.

DIFFERENTIAL AGING IN CHOLINERGIC NEURONS
The selection of LPl1 and R2 neurons was based on several criteria including their size, their reliable identifi cation in all preparations, their similar cell and transmitter phenotypes, as well as their functions. Size: both these cells are truly gigantic with an expected enormous yield of RNA. In fact, the R2 neuron shown in Figure 2B is the largest neuron ever found, measured or photographed so far, reaching 1.1 mm in a 1.3-kg animal! It appears that in this case the neuronal size had reached its potential limit since the neuron started to lose its spherical shape with the appearance of some obvious folds (convolutes) on its surface. It should be noted that we isolated the very same cell after photography (Figure 2B, see insert) then extracted a record 1.965 µg of RNA from this single neuron! Identifi cation: the CNS contains only a single copy of LPl1 and a single copy of R2. As a result, these neurons are among a few cell types with the easiest possible visual and a robust neuronal identifi cation and are in perfect lateral ganglionic positions for isolation purposes. These cells were recognized by Tauc and Hughes as symmetrical homologous, left (LPl1) and right (R2) motoneurons (Hughes and Tauc, 1961). Because of complex evolution and asymmetric fusion of ganglia in Aplysia development these cells are located in ganglia of different sizes (the larger abdominal ganglion is a result of fusion of three separate ganglia while the left pleural ganglion remains the same as the ancestral prototype).
These motoneurons control mucus release from body surfaces (Rayport et al., 1983;Ambron et al., 1988). We have cloned choline acetyltransferase ChAT from Aplysia (AY163563 , Table 1) and confi rmed its specifi c expression in these cells (not shown). Earlier biochemical measurements also suggested the presence of enzymatic ChAT activity in these cells and their cholinergic nature (Eisenstadt and Schwartz, 1975;Treistman and Schwartz, 1977). In addition these neurons co-express FMRFamide (Figure 4A), a neuropeptide with opioid like properties. Both neurons have largely symmetrical patterns of axonal branching and innervations (Hughes and Tauc, 1961;Kandel, 1976); a slight asymmetry might still exist since the animal does not have complete bilateral symmetry; there are asymmetrical reproductive, digestive and excretory organs and corresponding orifi ces. Known pharmacological and electrophysiological properties of these neurons are also nearly identical.
In summary, this pair of cells is a good model to perform an unbiased comparison of the transcriptional outputs from identifi ed neurons as well as their changes due to aging. Thus, we decided to compare single-cell gene expression patterns from individual homologous neurons from young and old animals using two types of experimental array experiments: direct and reference designs -as described in the method section. All together we performed gene expression profi ling from 22 individual neurons. For each experiment we analyzed expression data for ∼55,000 non-redundant transcripts (details of statistical analysis can be found in GEO GSE18783 and in the Section Materials and Methods). Figure 6 illustrates the importance of single-cell gene expression profi ling vs. profi ling from a relatively large population of heterogeneous cells (such as a single abdominal ganglion with about

FIGURE 6 | A comparison of gene expression profi ling between heterogenous ganglionic and single-cell samples from young and old animals. (A) Identifi cation of age-related transcripts using the abdominal ganglion (∼1000 cells). (B)
Identifi cation of age-related transcripts in the identifi ed cholinergic motoneuron R2. Both scatter plots show differential expression between the two samples of interest (log base 10). Differentially expressed transcripts from old animals are shown as red dots, and from young animals as green dots. Common transcripts (i.e., transcripts with similar expression levels in two samples) are removed. The scatter plot for a representative array in (A), when two abdominal ganglia were compared, indicates a smaller scale of detected differential expression vs. single-cell gene expression profi ling (B) under the same hybridization conditions. The single-neuron experiment also shows much higher diversity of differentially expressed transcripts (e.g., transcripts affected by aging). The direct comparison array experiment (B) identifi ed 3,855 differentially expressed genes in young and old R2 neurons (or 9% from the total number of tested transcripts: 2,291 transcripts were found to be differentially expressed in young R2 and 1,566 transcripts in old R2 at >2-fold change and a 5% FDR). These results also illustrate the importance and need for single cell analysis that takes into consideration the observed heterogeneity of even small neuronal samples (see text for details). All direct comparison arrays were performed with three biological replicates from different ganglia or single cells isolated from different animals. Differentially expressed transcripts for the single cell R2 experiments can be found at GEO Series accession number GSE18783 and Table S2 in Supplementary Material.

Moroz and Kohn
Differential aging in cholinergic neurons 1,000 neurons). Indeed, individual direct comparisons of "old" and "young" abdominal ganglia revealed about 1,300 differentially expressed transcripts with more than 2-fold change in expression. However, we could not achieve statistically reproducible identifi cation of differentially expressed genes when comparing abdominal ganglia from different animals of two age groups (statistical tests with ≥2-fold change and a 5% FDR). In contrast, when we performed direct single-cell comparisons from individual "old" R2 vs.
"young" R2 (i.e., neurons obtained from old and young animals) we detected statistically signifi cant changes in expression of 3855 transcripts ( Figure 6B and Table S2 in Supplementary Material at GEO Series accession number GSE18783) and these age-related differently expressed genes were independently supported in our reference design tests ( Figure 7A) and validated by QRT-PCR (Figure 8). Differentially expressed transcripts for the single cell R2 experiments can be found in GEO Series accession number GSE18783.

FIGURE 7 | Comparison of age-related changes in gene expression between two identifi ed cholinergic neurons LPl1 and R2. (A)
The schematic representation of the reference design microarray experiments to compare two different cell types R2 and LPL1 during the aging process. In these microarray tests, individual neurons were compared to the same reference CNS sample [see Materials and Methods, and legends for (C,D) below]. The individual circles represent single neurons (LPl1 -red tones; and R2 -blue tones) from young or old animals. (B) We found that only 58 neuronal transcripts (∼0.1%) are differentially expressed when LPl1 and R2 neurons are compared from young animals. In contrast, when the same cells were directly compared from old animals, we identifi ed 2508 differentially expressed transcripts (∼4.5%). This suggests that identifi ed cholinergic motoneurons are more similar to each other in younger animals than the same neuronal types in older animals. (C) The scope of detected age related transcripts in indentifi ed neurons from a reference design experiment as in (A). This histogram shows the total number of transcripts differentially affected by aging in both LPl1 (3930) and R2 (4440), as compared to a reference CNS sample, and a fraction of transcripts whose differential expressions were only found in one but not another neuronal type. For example, 2057 transcripts were found to be differentially expressed only in R2 neurons without signifi cant changes for the same transcripts in LPl1; correspondingly, we identify 1660 transcripts that changed their expression in aged LPl1 neurons but with a total of 58 transcripts differentially expressed. However, Old R2-Old LPl1 has >2500 transcripts differentially expressed suggesting neuronal transcriptomes change dramatically during aging. Second, it appears that the scope of age related changes is much more prominent in R2 neurons than in LPl1 neurons (compare blue asterisks for R2 and red asterisks for LPl1). The data summarized in (C,D) are based on the MANOVA analysis for both the single cell and aging effects within the reference type experiments as illustrated in (A). A >2-fold change and a ≤5% FDR differentially expressed transcripts for all the single cells were determined from combined AAA and DAA array data, see GEO series accession number GSE18783 and Tables S3 and S4 in Supplementary Material. Combined, these results suggest that the aging process in these neurons is not identical; differential subsets of genes uniquely and cells specifi cally change their expression as these neurons undergo normal aging. Examples of specifi c transcripts are listed in Table 2.
One of the proposed explanations of these results is the possibility of profound cell-specifi c changes in aging. For example, changes in opposite directions of the same groups of transcripts in different neurons will result in smaller or no detectable overall outcome because cell-specifi c changes will cancel each other in a highly heterogeneous sample. On the other hand, when uniquely identifi ed neurons were compared directly to each other, "contributions" from other neuronal groups are eliminated and observed age-related changes in gene expression can be unmasked. Figure 7B shows the scope of differential expression as a function of age between two neurons LPl1 and R2 from the reference design experiments and standard microarray statistical tests (n = 6 for R2 and n = 6 for LPl1, signifi cance tests with a 5% FDR and 2-fold change cut-off; see Materials and Methods and the access to the GEO 18783 project at NCBI site). We found that only 58 of 55,000 transcripts tested were differentially expressed between symmetrical young LPl1 and R2 neurons. In contrast, the same identifi ed neurons from old animals had 2058 differentially expressed transcripts (MANOVA and signifi cance tests with a 5% FDR and 2-fold change cut-off). Importantly, in both cases we identifi ed 2057 transcripts that are uniquely expressed in older R2 (but not in older LPl1) and 1660 transcripts in LPl1 from old animals but not in R2 from the same animals ( Figure 7C; the transcripts identifi ed as common can be found in Table S3 in Supplementary Material at GEO Series accession number GSE18783). About half of the genes signifi cantly affected by aging in these neurons appear to be occurring in one but not the other cholinergic neuron tested (Figure 7D; number of transcripts for each of the combinations are in Table S4 in Supplementary Material at GEO Series accession number GSE18783).
We interpret these data as cell-specifi c differential transcriptional changes underlying normal aging in these neurons. In other words, at the level of transcriptional outputs neurons do age differently.

A brief overview of transcriptional changes in cholinergic neurons
Although in this study we focused at neuron-specifi c differences in aging, the overall scope of detected transcriptional changes was quite extensive, totaling more than 5000 non-redundant transcripts with about 1000 annotated (see Table S5 in Supplementary Material, at GEO Series accession number GSE18783). These combined results allow identifi cation of an initial subset of gene categories that are shared among different neuronal classes. There are several evolutionarily conserved classes of genes encoding basic metabolic functions including apoptosis-and redox-dependent processes and insulin signaling. Many of these have already been recognized as important components of aging mechanisms in other model organisms (McCarroll et al., 2004).
Among Aplysia neuronal genes with up-regulated expression in aging are components of the mitochondrial respiratory chain (e.g., cyt c oxidase sub II), mitochondrial DNA polymerase involved in mitochondrial replication, catalase, V-ATPase, calmodulin, selected gap junction proteins, tumor necrosis factor receptors, transposases, DEAD/DEAH box helicases, prominin-1 (the stem cell marker), antistasin, cys-rich protein enriched in motor neurons, c-Jun N terminal kinase, selected 5-HT-and Glu-type receptors and plexin A4.
Among transcripts down-regulated in old animals are superoxide dismutase (SOD), insulin-like growth factor binding proteins, water channel protein -aquaporin-3, actin, epidermal growth factor receptor and pathway substrates, cell defense immune protein, and painless protein (transient receptor potential cation channel).
The transcripts showing highest differential expression in old R2 most notably are involved in systemic infl ammation such as proinfl ammatory cytokines, and tumor necrosis factor and its receptor (the same phenomenon also is observed in humans). Other examples of gene families up-regulated in "old" cholinergic neurons are oxidant stress related kinases, cell death proteins, gap junction proteins, fi ve K-channels, >30 receptors, and, importantly, several transposases and transposons. Also notable transcripts showing high differential expression in old R2 are Huntington's disease related transcripts (including huntingtin and huntingtin-interacting protein, and huntingtin associated transcripts).
In young R2 one of the most highly expressed transcripts is Msf1, encoding a predicted mitochondrial protein involved in neurodevelopment (Dee and Moffat, 2005). A novel family of mitochondrial proteins differentially expressed in young R2 is represented by homologs to Drosophila genes such as slmo and prel. Illustrative examples of gene families down-regulated in "old" cholinergic neurons were estrogen-regulated growth inhibitor, nexin and hyperpolarization-activated cAMP gated channel.
Finally, the list of differentially expressed "aging" genes includes RNA binding proteins such as major vault proteins that were up-regulated in "old" neurons whereas transcripts encoding Nogo protein (associated with synaptic functions and transport) were down-regulated.
In conclusion, this initial survey implies that transcriptomes of identifi ed neurons change dramatically during aging. As the next step, we focused upon two broad gene categories crucial for a large spectrum of modifi cations involved in control of neuronal identity and plasticity.

Validation of age-related differential expression for genes involved in cytoskeleton dynamics and chromatin remodeling
We used QRT-PCR to validate both cell specifi city and differential expression for selected genes. In all cases, we confi rmed the direction and scale of changes from our microarray experiments (Figure 8, Tables 2 and 3). As shown in Figure 8A the expression level of both presenilin and amyloid precursor protein (APP) mRNAs was higher in older animals (similar changes were also found for MCC neurons). Nevertheless, the expression of presenilin was more prominent in R2 (12-fold) than in LPl1 (1.5-fold) whereas changes in APP mRNAs were greater in LPl1 (5-fold) then in R2 (2-fold).
Remarkably, neuron specifi c changes in aging were also found for an inhibitor of apoptosis homolog protein (80-fold). This mRNA was not detectable (virtually not expressed) in young neurons (including MCC) but was highly abundant in R2 neurons from older animals (with a minor increase in LPl1). This may help explain the fact that senescent cells are resistant to apoptosis (Campisi, 2001).
In contrast, for transcripts associated with cytoskeleton remodeling we showed signifi cant reductions of their expression levels in older animals. For example, expression of gelsolin (actin capping protein involved in microtubule dynamics (Burtnick et al., 2001;Silacci et al., 2004;Spinardi and Witke, 2007), was less in all tested neuronal types (LPl1, R2 and MCC) from old animals ( Figure 8A).
Two key motor proteins are known to be involved in directional anterograde (kinesin) and retrograde (dynein) transport of RNAs, mitochondria and organelles in cells, including establishment and maintenance of neuronal polarity and long-term plasticity via targeted delivery of RNA cargo to support local protein synthesis at synaptic sites (Puthanveettil et al., 2008). The levels of expression for these motor proteins were signifi cantly lower in neurons obtained from old animals [i.e., for kinesin II (5-fold higher in young R2) and dyneins (35-fold higher in young R2)] suggesting alteration of the dynamic interaction between distant synapses and the cell nucleus in aging.
Importantly, we detected signifi cant changes in components associated with epigenetic modifi cations ( Figure 8B and Table 3). High expression levels of histone 3.3 (H3.3) are associated with active transcription (Lacoste and Almouzni, 2008;Jin et al., 2009). We found that H3.3 is slightly up-regulated in older R2 and MCC but down regulated in older LPl1. Signifi cant changes were also found for the histone H2B (20-fold higher in old R2, see also Figure 8B).
Cell-specifi c changes were the most pronounced in a distinct histone variant named histone Macro2A. So far histone-Macro 2A is described only for vertebrates but we were able to clone it from Aplysia (GB# EU104713). Out of three cell types tested (R2, LPl1 and serotonergic interneuron MCC) this histone variant is apparently highly expressed only in LPl1 neurons and its level is signifi cantly reduced in neurons from aged animals ( Figure 8B). In Aplysia the functional role of histone-macro2A is unknown. However, microarray-based analysis in mammals uncovered occupancy of macroH2A variants at many genes encoding key regulators of development and cell fate decisions. Specifi cally, macroH2A contributes, as an epigenetic mark, to the fi ne-tuning of temporal activation of HOXA cluster genes and regulation of the gene expression program during cellular and neuronal differentiation (Chang et al., 2008;Buschbeck et al., 2009;Timinszky et al., 2009). Age-related changes in Aplysia neurons were also observed for histone modifying proteins. For example transcripts for putative histone acetyltransferase HPA2/1 were found to be less expressed in R2 (11-fold) suggesting suppression of transcription for a number of genes.
One of the most remarkable features of Aplysia is the presence of DNA (cytosine-5)-methyltransferase (DNMT-1) and subsequent DNA methylation. Apparently this process was lost in both Drosophila and C. elegans (Suzuki and Bird, 2008). Here we found a 3-fold greater expression level for DNMT-1 in R2 from older compared to younger animals, further suggesting that DNA methylation transcriptional repression might play an important role in the control of genome-wide chromatin reorganization in aging. Consistent with this idea, we discovered that the transcriptional repressor MBD2 (methyl-CpG binding domain protein 2, see Dhasarathy and Wade, 2008;Tiwari et al., 2008;Martin Caballero et al., 2009) is also more abundant in R2 from old animals.
In summary, our data suggest that epigenetic mechanisms can be crucial for normal aging and be important players responsible for neuron-specifi c changes accumulated during this process.

DISCUSSION AND FUTURE DIRECTIONS DIFFERENTIAL AGING IN NEURONS AS A WIDESPREAD PHENOMENON
Functional, morphological and chemical heterogeneity of neurons is the most fundamental and distinct feature of virtually any neuronal circuit or nervous system. The Allan Brain Atlas project (Lein et al., 2007) added a new dimension to this picture. It revealed unprecedented complexity of genome-scale variations across cell populations suggesting that the majority of neurons in our nervous system are unique cells in terms of their genomic organization, connectivity and specifi c functions. Numerous clinical and experimental studies dealing with neurodegenerative diseases also indicate prominent cell-type specifi city in vulnerability or resistance of various cell populations. This view can be reasonably applied to mechanisms of aging and age-related diseases in our brain.
In this study, using direct single-cell genomic profi ling, we wanted to test the hypothesis that different identifi ed neurons can age differently even if they belong to the same class of motoneurons. For comparison we selected one pair of identifi ed cholinergic motoneurons that control mucus release from the body surface as a part of the defense refl ex but that also contribute to a tonic steady state mucus release under basal conditions. In fact, these are two very similar homologous cells innervating the left (LPl1) and right (R2) part of the body respectively (although they do not have absolutely identical innervation fi elds and development).
One of the major conclusions of this study is the evidence that the aging process in these neurons is not identical; different subsets of genes uniquely and cell specifi cally change their expression as these neurons undergo normal aging.
Our initial confi rmation that different neurons do age differently is not a very surprising outcome per se considering the enormous molecular heterogeneity within neuronal populations. What is more remarkable is the scale of differential changes in gene expression profi les even between two very closely related neurons such as LPl1 and R2 (Figure 7). Indeed, the age related differences between two symmetrical cholinergic neurons are more complex than we initially expected. It appears that as neurons age they become as different from their counterpart young neuron as a different cell type. The scale of 4-7% differences in differently expressed genes between young and old neurons of the same type (vs. ∼0.1% between two different young neurons) is comparable to differences between sensory and motor neurons found in our previous microarray tests .
In our analysis and QRT-PCR validation tests (Figure 8) we primarily focused on highly abundant transcripts. However, we predict that the extent of cell-cell differences could be signifi cantly greater because microarrays have a lot of constraints, including limited sensitivity and diffi culties in detection of low abundance transcripts (and even transcripts with medium abundance). Expression profi ling using the direct sequencing strategy (or RNA-Seq - Tang et al., 2009;Wang et al., 2009) would be more effi cient than microarrays to further validate these hypotheses and quantify transcripts within the whole dynamic range of their abundance. These approaches are feasible for single Aplysia neurons (Kohn et al., 2008). Massive parallel sequencing technologies (e.g., 454, Illumina, SOLID and other established and emerging platforms) will be essential for direct and unbiased analysis of the operation of the entire cellular genome across different cell types in aging (sequencing and annotation of the Aplysia genome is in progress now).

EPIGENETIC LANDSCAPE DURING AGING AND CELL FATE SPECIFICATION
An organism's longevity and aging is caused by numerous interacting factors such as nuclear and mitochondrial genome mutations, shortened telomeres, oxidative damage to DNA and other macromolecules, senescence, apoptosis and many more (Masoro and Austad, 2006;Campisi and d'Adda di Fagagna, 2007;Vijg, 2007;Vijg and Campisi, 2008). We show here that homologous neurons age differently, with prominent changes in epigenetic components during their aging process.
An analogy to the classical Waddington's epigenetic landscape could be applied as a conceptual visualization for differential aging in neurons (Figure 9). In other words, differential aging can be comparable to distinct cell fate specifi cation with non identical molecular mechanisms. Under this hypothesis, epigenetic mechanisms serve as master regulators controlling genome-wide chromatin remodeling and transcription. On the other hand, the same components acting in different neurons would also operate within specifi c chromosome territories and gene regulatory regions to maintain neuronal homeostasis and cell specifi city.
Similarly, large-scale analysis of molecular markers involved in cell lineage development in C. elegans confi rmed that the same cell fate can be achieved using different gene regulatory pathways and mechanisms (Liu et al., 2009).
Our data clearly suggest that all three components of chromatin remodeling machinery -different histone variants (e.g., H2B, histone-macro2A), histone modifying enzymes and DNA methylation components (DNMT1 and associated methyl-CpG binding domain protein -MBD2) change as a function of aging. Importantly, all of them do show some degree of cell specifi city suggesting that these changes might underlie differential expression of thousands of genes in identifi ed neurons. Finally, our pilot in situ hybridization data indicate differential expression in the CNS of Aplysia DNMT1 (not shown, Kohn et al., 2009), suggesting that the tonic level of DNA methylation can also be an important mechanism for controlling cellular phenotypes in neural circuits. Ultimately, it would be not surprising to fi nd out that each neuron has its own epigenome.
Even though each neuron tested (e.g., MCC, LPl1 and R2) shows profound transcriptional changes during aging, most of them also show remarkable cell-specificity both in their expression patterns and in directional changes in young and mature animals. These results imply that global and possibly stochastic molecular events in aging could be an important factor of permanent and irreversible changes in both selective gene expression patterns and the entire transcriptional output of neurons. Some of our unusual findings that need further analysis and validation are the potential involvement of transposons and transposases in aging. Although these categories of regulators are not annotated in Aplysia, a number of expressed transcripts belong to recognized evolutionarily conserved subgroups of mobile genomic elements and can be responsible for irreversible changes in DNA, genomic instability in aging (Busuttil et al., 2004) and senescence.

CYTOSKELETON AND SYNAPTIC FUNCTION IN NEURONAL AGING
A second large category of transcripts consisting of regulators of local protein synthesis and regional transport was also found to be differentially expressed in aging. Indeed, single-neuron microarray tests suggest that both kinesin and dynein expression are down-regulated in "old" cholinergic neurons. These two proteins are major cargo complexes mediating anterograde (kinesin) or retrograde (dynein) transport of mRNAs and organelles. Thus, their suppression in aging should have signifi cant functional consequences for synaptic function (e.g., likely leading to a decrease of synaptic plasticity, maintenance of neuronal morphology and polarized connections, see Puthanveettil et al., 2008). Our data also revealed that changes in antisense profi ling can be as complex as the expression profi le of any single neuron and some of them can be neuron-type specifi c. This broad class of transcripts that include microRNAs (Rajasethupathy et al., 2009) and large non-coding RNAs such as reported here 451 (GB#AF401632.1, Table 2) presents unique opportunities in analysis of regulatory mechanisms especially as local regulators of transcription and translation.
Previous studies indicated that changes in transmitter (5-HT, DA, ACh) synthesis or degradation, 5-HT and DA receptor functions, changes in presynaptic proteins and some neuropeptides (FMRFamide and egg-laying hormone) can be important mechanisms of neuronal aging processes in Aplysia (Kindy et al., 1991;Peretz and Srivatsan, 1996;Flinn et al., 1997;Southall et al., 1997;Ono et al., 1998;Chandhoke et al., 2001). Here, we partially confi rm these earlier observations (e.g., differential expression of FMRFamide in older animals) as well as expand these data to many other peptides such as buccalin (GB# P20481 (4-to 5-fold less in old animals) and ionotropic glutamate receptors (the level of iGluR5 is less in old animals).
Finally, it should be noted that in Aplysia and related gastropod molluscs the neurons increase in size as the animals grow in body size. Therefore, some observed changes in transcriptional output could be related to changes in neuron sizes and refl ect changes in the surface to volume ratios. However, we took this factor into consideration for this study about differential aging. Specifi cally, we have compared aging of two neurons which grow at similar rates and reach comparable sizes as they age. Thus, we feel that it is unlikely that the observed differences in gene expression patterns between these two neurons as they age can be attributed to neuron sizes.
Nevertheless, the precise relationships between transcriptome, aging and cell growth rate are currently unknown, but this analysis can be performed in the future when multiple cell types of different sizes would be compared. Apparently, the growth rate is not identical for different neuronal classes since even an adult Aplysia has many neurons of small sizes (e.g., sensory cells). Furthermore, separate studies can be performed by comparing various identifi ed neurons from mature vs. very old animals where neurons have comparable sizes between these two stages.

APLYSIA AS AN EMERGING MODEL FOR THE CELL AND GENOMIC BIOLOGY OF AGING
To the best of our knowledge, this is the fi rst genome-scale profi le of identifi ed neurons as a function of aging. At this moment, it is diffi cult to say whether the changes we observed for Aplysia would be within the same range for human or other mammalian neurons. We hypothesize that this would be the case. The lack of identifi ed neurons and anatomical diffi culties in performing such FIGURE 9 | Waddington type landscape superposed with the distinct epigenetic trajectories for different neuronal types as a function of aging. The landscape diagram is modifi ed from Waddington, C. H., 1956 (Principles of Embryology, op. cit., p. 412). Following Waddington's visual schematics, the ball represents a neuronal fate. The valleys are the different fates a given neuron might roll into. At the beginning of its journey, development is plastic, and a cell can have many fates. However, as development and aging proceeds, certain molecular events occur randomly and this can lead to different underlying molecular phenotypes and decisions that cannot be reversed.
analyses within highly diverse neuronal populations in vertebrate brains are major obstacles in the fi eld. Although it is expected that such experiments would be feasible in the near future (e.g., using transgenic lines with neuron-type specifi c fl uorescent markers), Aplysia offers unique and unprecedented opportunities for single-cell genomic and epigenetic studies of aging and age related memory loss in particular. The development of Aplysia (Kriegstein et al., 1974;Kriegstein, 1977;Kandel, 1979;Heyland and Moroz, 2006) and phenomenology of aging as well as age-related changes in Aplysia behavior and physiology, in particular on the gill withdrawal refl ex and synaptic plasticity (Peretz et al., 1982Peretz and Romanenko, 1986;Peretz, 1988;Peretz andSrivatsan, 1992, 1996) are well described, providing a solid background for this model.
The molecular and genomic biology underlying elementary cognitive functions and plasticity can be studied in Aplysia with resolution that is diffi cult to achieve elsewhere because identifi ed neurons, synapses and relatively simpler memory forming circuits have been well characterized (Kandel, 2001;Hawkins et al., 2006;Bailey and Kandel, 2008). The accessibility and size of these neurons is the perfect experimental platform for development and integration of genomic/epigenomic profi ling of biochemical and physiological functions at a single cell level. Both single cells and nuclei from specifi c cells can be isolated within minutes. Moreover, <1% of the cytoplasm from a single cell is all that is needed to directly characterize and quantify hundreds of intracellular metabolites, signal molecules and various peptides using capillary electrophoresis and mass spectrometry Li et al., 2000;Li and Sweedler, 2008;Lapainis et al., 2009).
In addition, we showed that Aplysia has many common and evolutionarily conserved age-related gene families that are associated with antioxidant defense, stress response, mitochondrial function, insulin signaling and infl ammation pathways, as well as gene homologs of neurodegenerative diseases including Alzheimer's, Huntingtin and Parkinson's as well as pain Walters and Moroz, 2009). Differential age-related expression of amyloid peptide precursor (in LPl1) and huntingtin (in R2) suggests that it would be possible to develop and study molecular, cellular and circuit mechanisms in Aplysia at the level of uniquely identifi ed neurons.
Obviously, a paradigm shift is required to study aging in specifi c neurons with a prospect to perform nearly complete unbiased single-neuron genomic analysis with identifi cation and quantifi cation of all cellular RNAs. Such a new approach, with the ability to perform metabolomic and proteomic profi ling from individual neurons and to access identifi ed neurons with different dynamics of natural aging processes, opens a broad spectrum of opportunities to study and understand why some neurons can age at different rates, degenerate or die whereas others can be virtually unaffected by aging. By following such an integrative molecular portrait of individually identifi ed neurons during the aging process we see the virtually unmatched power of Aplysia as the emerging model in the biology of aging.
Conceptually, the use of Aplysia as a novel genomic model can bring integrative genomic, molecular and physiological analyses to the level of single cells and cell compartments. At the same time it will also bridge all levels of system organization during the organism's aging: from the understanding of dynamic reorganization of the cellular genome to specifi c neural circuits and then to behaviors, learning and memories. Although we are developing the new marine model for analysis of aging mechanisms, the proposed approaches can be implemented in other organisms in both aging research and genome biology of cognitive functions.