Quantitative analysis of the detergent-insoluble brain proteome in frontotemporal lobar degeneration using SILAC internal standards

Case selection and protein extraction

Cases included in this study all had a diagnosis of FTLD, as determined by extensive clinical assessment and neuropathological characterization based on well established consensus criteria at the Emory Alzheimer’s Disease Research Center [19, 20]. The cases characterized as FTLD exhibited phosphorylated TDP-43 positive, tau- and α-synuclein-negative, cytoplasmic inclusions in neurons of both the superficial frontal cortex and dentate gyrus of the hippocampus. Additionally, these cases did not meet criteria for the neuropathological diagnosis of AD (CERAD [21] or NIA-Reagan [22]) or Lewy Body disease [23], nor did they show any tau pathology consistent with a tauopathy [24, 25]. In total, frozen frontal cortex from twelve FTLD and eleven control cases (Supplemental Table S1) were homogenized and detergent-insoluble protein extraction were performed by differential ultracentrifugation as described previously [26].

Stable isotope labeling with amino acids in cell culture (SILAC) and LC-MS/MS

HEK293 cells (~1 x 10⁸) were cultured in DMEM (deficient in L-Lysine and L-Arginine) supplemented with 2% dialyzed fetal calf serum (Invitrogen) as described [27]. For stable isotopic labeling, heavy forms L-Arginine (Arg10; ¹³C₆¹⁵N₄) and L-Lysine (Lys8; ¹³C₆¹⁵N₂) were added (Cambridge Isotope Laboratories) to a final concentration of 0.26 mM. Excess proline was added at 200 mg/L to block arginine to proline conversion [28]. Cells were washed twice and collected in ice cold phosphate buffered saline (PBS) and then lysed in ice cold cell lysis buffer (50mM Tris·HCl pH 7.4, 1mM EDTA, 150 mM NaCl, 1% (v/v) Triton X-100, 1% (w/v) sodium deoxycholate, 0.1% w/v SDS). Complete protease cocktail inhibitor (Roche, Indianapolis, IN) was added to all buffers immediately prior to use. Before SDS-PAGE, 10 μg of SILAC labeled HEK293 whole cell lysate was added to 10 μg of detergent-insoluble fractions from four pooled control and four pooled FTLD extracts, respectively, matched as closely as possible by age and post-mortem interval (Supplemental Table S1). The mixed (light and heavy) samples were reduced with 10 mM dithiothreitol (DTT), and resolved on a 10% polyacrylamide SDS gel. After staining with Coomassie G-250, each gel lane was cut into five gel pieces according to molecular weight (A-E; ≤ 25, 25–40,40–60, 60–100 and ≥ 100 kDa, respectively) and subjected to in-gel digestion (12.5 ng/μl trypsin). Extracted peptides were loaded onto a C₁₈ column (75 μm i.d., 10 cm long, ~300 nl/min flow rate, 3 μM resin from Michrom Bioresources, Auburn, CA) and eluted during a 10–30% gradient (Buffer A: 0.4% acetic acid, 0.005% heptafluorobutyric acid, and 5% AcN; Buffer B: 0.4% acetic acid, 0.005% heptafluorobutyric acid, and 95% AcN). The eluted peptides were detected by Orbitrap (300–1600 m/z, 1,000,000 automatic gating control (AGC) target, 1,000 ms maximum ion time, resolution 30,000). MS/MS scans in an LTQ linear-ion trap mass spectrometer (2 m/z isolation width, 35% collision energy, 5,000 AGC target, 150 ms maximum ion time) (Thermo Finnigan, San Jose, CA) were acquired by data-dependent acquisition. All data were converted from raw files to the .dta format using ExtractMS version 2.0 (Thermo Electron) and searched against human reference database downloaded from the National Center for Biotechnology Information (November 19, 2008) using the SEQUEST Sorcerer algorithm (version 3.11, SAGE-N). Searching parameters included mass tolerance of precursor ions (±50 ppm) and product ion (±0.5 m/z), semi-tryptic restriction, with a dynamic mass shift for oxidized Met (+15.9949), Lys (+8.01420 for ¹³C₆¹⁵N₂) and Arg (+10.00827 for ¹³C₆¹⁵N₄), four maximal modification sites and a maximum of two missed cleavages. Only b and y ions were considered during the database match. To evaluate false discovery rate (FDR), all original protein sequences were reversed to generate a decoy database that was concatenated to the original database (77,764 entries) [29]. The FDR was estimated by the number of decoy matches (nd) and total number of assigned matches (nt). FDR = 2*nd/nt, assuming mismatches in the original database were the same as in the decoy database. To remove false positive matches, assigned peptides were grouped by a combination of trypticity (fully and semi) and precursor ion-charge state. Each group was first filtered by mass accuracy (10 ppm for high-resolution MS), and by dynamically increasing correlation coefficient (Xcorr, minimum 1.0) and ΔCn (minimum 0.05) values to reduce protein FDR to less than 1 percent. All MS/MS spectra for proteins identified by a single peptide were manually inspected as described previously [30]. If peptides were shared by multiple members of a protein family, the matched members were clustered into a single group. On the basis of the principle of parsimony, the group was represented by the protein with greatest number of assigned peptides.

Peptide quantification and bioinformatics

Quantitative pair-wise comparison of control and FTLD samples was performed in technical replicate according to previously reported methods using in-house software [15, 31, 32]. (i) Ion extraction from MS scans: The ion currents for identified peptides were extracted in MS survey scans of high-resolution (30,000), based on the isotopic ion selected for MS/MS sequencing. A number of features were derived, including precursor m/z, charge state, retention time, ion peak width, height and noise level. The global noise level (GNL) was derived by averaging signal intensity of all ions in the MS scan after removing signal ion outliers that were at least two standard deviations away from the mean. The intensity of ions was presented by the peak height and normalized according to the noise intensity under the assumption that the noise level of MS scans reflects, at least partially, variable ionization efficiency. (ii) Ion matching among light, and heavy peptide pairs was allowed with a mass tolerance of 10 ppm. If a sequenced peptide could not be matched to a signal above GNL, we estimated that the maximum ion current for undetected signal was equal to the local noise level (LNL) of ions within the 10 ppm window (LNL ≥ 0.5). (iii) Data integration: The ratio of every peptide pair (light/heavy) was transformed into logarithmic (log₂) values and the detection of outliers determined using Dixon’s Q test [33]. If a peptide was identified in multiple MW fractions the intensities for the peptide pair (light and heavy) were summed to give a single cumulative peptide ratio (light/heavy). The averaged log₂ ratio over all peptides for a particular protein was used to determine the protein ratio and a standard error (S.E.). The results are summarized in Supplementary Tables S2 and S3 with the number of peptides quantified for every protein and the individual peptide ratios for control and FTLD (Supplementary Tables S5-S8) from each technical replicate. (iv) Data normalization: According to the null hypothesis, the histogram of the difference (FTLD-Control) between all shared protein log₂ ratios was fitted to a Gaussian distribution to evaluate systematic bias according to the mean and biological variation based on standard deviation using Graphpad Prism (ver. 5.0, GraphPad Software, La Jolla, CA). The data were then normalized by subtracting the mean from every protein ratio. Proteins were considered enriched if the brain/SILAC log₂ values for control and FTLD peptide mixtures (within a replicate) had signal/noise (S/N) greater than 10 and were quantified by two or more peptides. Analysis of bimodal distribution in Control- and FTLD- SILAC (HEK293) mixtures was performed using multi-peak fit package on Igor (version 6.0, WaveMetrics, Inc., Lake Oswego, OR). Finally, the quantified proteins were manually examined with respect to MS/MS assignment, ion peak matching and ion intensity. To confirm that brain-specific proteins are a major component of the unlabeled and unpaired proteins present in the brain/HEK293 SILAC protein mixture, we performed Ingenuity Pathway Analysis IPA (Ingenuity® Systems, www.ingenuity.com). NCBI RefSeq identifiers for proteins identified with a SILAC light/heavy log₂ ratio above 3 were submitted to IPA. The identifiers mapped to 604 unique genes, and the top disease-associated pathway was noted as neurological disease, involving 50 percent (301) of the genes. In addition, the top physiological function represented was nervous system development and function, involving 30 percent (179) of the genes. All nervous system pathway lists of genes represented were exported with p-values and genes in each list were counted, to produce a chart of the number of proteins identified as brain enriched and involved in the top physiological function represented.

Immunoblotting

Immunoblotting was performed according to standard procedures as reported previously [15]. Briefly, samples in Laemmli sample buffer were resolved by SDS-PAGE (12% w/v acrylamide) before an overnight transfer to Immobilon-P membranes (Millipore, Bedford, MA). These were blocked with TBS (Tris-buffered saline) plus blocking buffer (USB Corporation, Cleveland, OH) at room temperature for 90 min and probed with primary antibodies in TBS with 0.1% Tween-20 plus blocking buffer overnight at 4°C. The following day, membranes were rinsed and incubated with secondary antibodies conjugated to fluorophores (Molecular Probes/Invitrogen) for one hour at room temperature. Images were captured using an Odyssey Image Station (LiCor, Lincoln, NE) and band intensities were quantified using Scion Image. Statistical analysis was performed using Student’s t test for independent samples. Antibodies included: Mouse anti-CNDP2 (R&D Systems, MAB3560), rabbit anti-PSF (Abcam, ab38148), rabbit anti-TDP-43 (ProteinTech, 10782-1-AP), rabbit anti-cofilin-1 (Santa Cruz) rabbit anti-RHOGDI-alpha (ProteinTech,10509-1-Ig), rabbit anti-PGP9.5 (i.e, UCHL1) Chemicon International, Temecula CA), mouse anti-CNP (Chemicon International, Temecula CA), rabbit anti-PTB (Invitrogen) and goat anti-TPI1 (Abcam, ab28760).

Immunocytochemistry

The GFP-PSF wild-type construct was kindly provided by James G. Patton (Vanderbilt University) and characterized as previously described [41]. HEK293 cells were cultured in Dulbecco’s Modified Eagle’s medium (DMEM) (Cambrex, Walkersville MD), and supplemented with 10% fetal bovine serum (Gibco, Grand Island NY) and 1% penicillin-streptomycin (Cambrex). Cells were grown in a humidified 5% CO₂ environment at 37°C, and prepared for transfection (Lipofectamine 2000) by plating on Matrigel-coated coverslips. Immunocytochemistry was performed 48 hours after transfection as previously described [42]. Images were captured with a 1μm optical thickness for subsequent analysis on a Zeiss LSM 510 confocal microscope (Zeiss, Thornwood NY).

Primary cell culture

Primary cultured rat cortical neurons were raised using a previously described procedure [43]. Briefly, cortices of E16 Sprague-Dawley pups were minced and gently agitated to obtain dissociated cells in DMEM containing 5% fetal calf serum and 5% horse serum. After washing, cells were resuspended and seeded onto poly-L-Lysine-coated plates and incubated at 37°C in 5% CO₂. After 1 hr, cells were re-fed with Neurobasal medium containing B-27 supplement (Invitrogen) and incubated for 5 days before harvest. Primary cultures of oligodendroglia were raised using a well-established procedure [44]. Briefly, mixed glial cultures were prepared from the brains of neonatal Sprague-Dawley rats (P2), plated on 75 cm² flasks and allowed to reach confluence in DMEM containing 12% fetal bovine serum (FBS). Oligodendroglia progenitor cells (OPCs) were removed by shaking at 180 rpm over night and further purified from microglia and astrocytes that quickly attach to uncoated Falcon dishes. The floating OPCs were isolated and plated onto poly-lysine-coated dishes and maintained in a defined serum-free medium with platelet-derived growth factor and fibroblast growth factor. Differentiation was induced by removing growth factors and mature oligodendrocytes were obtained after 5 days in the differentiation medium lacking growth factors. After the shaking step defined above, the confluent glia layer contained more than 90% pure astrocytes. The cells were re-plated into medium containing 10% serum and harvested 24 hrs later. Primary cultures of microglia were raised as previously described [45]. Briefly, the brains of 1-day-old neonatal Sprague-Dawley rats were minced and cells dissociated by agitation before passing through a nylon membrane (pore size 75μm). Mixed glia culture was raised in RPMI-1640 containing 10% FBS, 2 mM glutamine, and 1% penicillin-streptomycin on poly-L-lysine coated flasks and allowed to reach confluence after 14 days. Microglia were isolated by gently shaking the mixed glia culture for 2 hr at 150 rpm at 37°C and plated on uncoated Costar culture dishes to remove non-adherent astrocytes. The adherent microglia was further cultured for 24 hr before harvest.

Immunohistochemistry

Paraffin-embedded sections from post-mortem human frontal cortex (8 μm thick) were deparaffinized and microwaved in citrate buffer (10 mM, pH 6) for 5 minutes. After cooling to room temperature, sections were rinsed and endogenous peroxidase activity was blocked with 3% hydrogen peroxide at 40^oC. Sections were then incubated with normal horse serum for 15 minutes at 40°C, followed by rabbit anti-PSF primary antibody (Abcam; 1:200), rabbit anti-TDP-43 (1:1000, ProteinTech), mouse anti-Olig2 (1:200 Millipore) or mouse anti-phosphorylated TDP-43 (Ser409/410) antibody (1:8000, CosmoBio) diluted in 1% BSA for 3,3′-diaminobenzoic acid (DAB) development or 0.25% (v/v) Brij-35, 100 mM NaCl, 50 mM MgCl₂,100 mM Tris-HCl, pH 7.5 for immunofluorescence, overnight at 4°C. The following day, sections were incubated with biotinylated secondary antibody for 30 minutes at 37°C, followed by avidin-biotin peroxidase complex (Vector Laboratories) for 60 minutes at 37°C. DAB was used as the chromogen (for color development) and was followed with hematoxylin nuclear counterstain. For pre-adsorption studies of PSF, primary antibody was pre-incubated with 100x molar excess of the immunizing peptide (Abcam) overnight at 4°C before incubation with the tissue. For immunofluorescence in paraffin embedded tissue, Alexa-488 and Alexa-fluor 546 were used as secondary antibodies (Jackson ImmunoResearch, Suffolk, UK). Free floating brain sections (50 μM) were prepared with a freezing microtome (Microm, Heidelberg, Germany) from FTLD brain blocks of the frontal cortex. Sections were subsequently incubated with serum followed by rabbit anti-PSF and mouse anti-GFAP (Millipore) primary antibodies overnight at 4°C. After washes with Tris-buffered saline, sections were incubated with fluorescent secondary antibodies (Alexa-488 and Alexa-fluor 546) for 1 hour at 4°C. Staining was visualized using an Olympus BX51 florescent/bright field microscope or confocal microscopy as described above.

Analysis of the detergent-insoluble proteome from FTLD and control brain

To identify and quantify detergent-insoluble proteins in FTLD cases (Fig. 1A), we examined post-mortem brain tissue samples using SILAC internal standards. In this strategy, four cases of control and four cases of FTLD frontal cortex homogenates were pooled by diagnosis and matched for age and sex (Supplemental Table 1). Samples were serially extracted with buffers of increasing stringency: First by sarkosyl detergent-containing buffer and then by urea to enrich for detergent-insoluble proteins [26, 46]. By pooling the urea solubilized proteins from our cases prior to discovery-mode proteomic experiments, the between-subject variability and pathophysiological heterogeneity as well as the variability in the efficiency of detergent extraction can be diluted [26, 47]. For quantitative proteomics, whole cell lysate from SILAC labeled HEK293 cells was mixed equally with detergent-insoluble fraction pools derived from either control or FTLD cases (Figs. 1B and 2A). Shared SILAC-labeled HEK293 proteins served as internal standards for software based measurement of protein abundance between control and FTLD samples (Fig. 2B). Each protein fraction was resolved by SDS-PAGE and the brain/SILAC (light/heavy) mixtures of control and FTLD samples were excised into five gel slices, trypsin digested, and analyzed via liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) on an Orbitrap mass spectrometer (Fig. 2A and C).

Quantitative analysis of shared proteins in the brain/SILAC mixed proteome versus label-free quantitation of brain-derived proteins

Quantitative pair-wise comparison in the brain/SILAC mixed samples was carried out in technical replicate as described in the materials and methods section. In control/SILAC sample, protein log₂ values (light/heavy) were plotted as a histogram and a bimodal distribution was observed (Fig. 3A). A similar bimodal pattern of distribution was observed in the FTLD/SILAC mixture (Supplemental Figure 1). Analysis of all pooled (FTLD/SILAC and control/SILAC) technical replicate LC-MS/MS data showed higher variation at protein log₂ values greater than 3.0 and less than −2.4 (Fig. 4A and B). Thus, the heavy labeled HEK293 internal standard greatly reduced the standard error in the proteomic quantification within this range (log₂ −2.4 to 3.0) when compared to those proteins quantified outside this range (log₂ <−2.4 or >3.0). It should be noted that we included both semi-tryptic and tryptic peptides in protein quantification. As expected, the control brain sample had a significant level of proteolysis compared to HEK293 cell standard, which may be a result of both brain specific in vivo cleavage events as well as post-mortem interval (Supplemental Figure 2). Importantly, the inclusion of semi-tryptic peptides did not overall negatively impact the accuracy of quantification of most proteins reported because they represent a small proportion (<8%) of total peptides quantified. Proteins with log₂ values below −2.4 mainly represented HEK293-specific or -enriched proteins (i.e. having much higher levels in the SILAC internal standard compared to brain). For example, peptides from SET translocation protein (myeloid leukemia-associated) had an average log₂ value of −5.17 and were more intense in the heavy labeled HEK293 cells than in unlabeled (light) native brain (Fig. 3B). Conversely, those proteins having log₂ values greater than 3.0 mainly represented brain specific or enriched proteins (i.e. having a much higher level in brain compared to the SILAC internal standard). These included the neuron specific protein microtubule-associated protein-2 (MAP2; log₂ = 4.25) (Fig. 3D). For proteins with log₂ values between the lower and upper limits (−2.4 to 3.0), such as ubiquitin activating enzyme 1 (UBA1), the intensity of matching peptides was sufficient for calculating the light/heavy ratio (Figs. 3C and 4). In the second replicate control brain proteome 34.0% (1032/3032) and 10.7% (323/3032) can be considered proteins outside the upper and lower limits (−2.4 to 3.0) of quantification. This highlights an important limitation of the HEK293 generated SILAC internal standard in that of all brain proteins identified only approximately 62% (1677/2709) are shared with brain, likely proteins which are found across most cell types, rather than proteins specific to brain. We confirmed this by performing Ingenuity Pathway Analysis using the 604 gene products identified specifically as unlabeled peptides (log₂ brain/SILAC ratio > 3). This analysis confirmed that 50 percent of these proteins can be linked to neurological disease. Further, 30 percent of the gene products identified, but not well quantified by SILAC standards are involved in nervous system development and function (Supplemental Figure 3). Nonetheless, we hypothesized that we could validate changes and brain-specific function of one or more of the shared or “universal,” SILAC-quantified, changing proteins as relevant to FTLD pathogenesis, as described in below results sections.

Here we define the criteria for consideration of brain/HEK293 shared proteins, and briefly extend analysis to rescue identifications of changing brain-specific proteins in FTLD. A histogram of the difference (FTLD-Control) between all shared protein log₂ values was fitted to a Gaussian distribution to evaluate systematic bias according to the mean and variation based on the standard deviation (SD) for the replicates (R1 and R2) individually (Fig. 5B). The SD in R1 and R2 was calculated to be 1.0 and 0.90, respectively. The difference in log₂ protein values (FTLD-Control) in both replicates were normalized by subtracting the experimental mean, 0.32 in R1 and 0.40 in R2. As a filtering criterion, FTLD proteins with potentially increased abundance that fell outside the 90% confidence interval (1.64 SD away from the mean, p< 0.05 in a one-tailed analysis) were considered as a subgroup of interest. In order to reduce the variability on protein measurements largely based on non-paired peptides, we required that the log₂ ratio in at least one of the two samples (control and FTLD) for either replicate fall within the log₂ range of −2.4 to 3.0. Two more criteria were required of the proteins appearing on the final table: i) the average log₂ value from R1 and R2 must indicate at least 2 fold enrichment (average log₂ >1.0) and ii) S.E between replicates ≤0.5. In total, twenty-one proteins met these criteria and were considered enriched in FTLD using the SILAC approach. These proteins are classified according to their cellular location and function (Table 1). The difference (FTLD-Control) between log₂ values for all proteins in R1 and R2 quantified is provided in Supplementary Tables S2 and S3.

To assess whether the protein fold changes we observed using SILAC internal standard were also reflected in a label-free quantitation approach via spectral counts (SC), and to potentially rescue the identification of FTLD changed proteins that are brain-specific, we performed G-test analysis of SC only derived from light labeled peptides in the FTLD and control samples (Replicate 1 and 2). Of the enriched proteins reported in Table 1 (quantified using SILAC standards), 11 of the 21 (52%) were also found significantly enriched by SC (p<0.01). Spectral counting is based on the observation that more abundant peptides will be selected for fragmentation and will produce a higher number of MS/MS spectra, and therefore, SC should be proportional to protein amount in data-dependent acquisition [48]. However, it is not surprising that proteins identified by only a few unlabeled peptides with low total SCs were not considered significant by G-test. For example RHOGDI-alpha, calreticulin and cathepsin D, which were considered enriched using SILAC internal standards, were sequenced with too low a frequency in the mass spectrometer (<15 total SCs in both control and FTLD) to be considered significant by G-test. Importantly, SC analysis did reveal several brain-specific proteins as being significantly enriched in FTLD that were outside our upper limit of detection (log₂ >3) using SILAC internal standard (Table 2). These proteins included brain creatine kinase (CKB), glial fibrillary acidic protein (GFAP), secernin (SCRN) and peptidylprolyl isomerase A (PPIA), which lacked significant signal from internal peptide standards for accurate quantification. All proteins, normalized spectral count ratios, and p values can be viewed in Supplementary Table S4. Thus, complementary quantitative approaches such as SILAC internal standards and spectral counting, respectively, enhance the ability to quantify low abundant proteins with shared brain peptides, or brain specific and brain-enriched proteins albeit with lower sensitivity. The remainder of this study focuses on the proteins identified as changing via SILAC quantitation.

Immunoblotting to confirm enrichment of insoluble proteins in FTLD brain tissue

Our primary goal was to employ an unbiased approach to identify disease-associated components in the detergent-insoluble FTLD proteome, which can include inclusion proteins among other insoluble proteome cellular components. To validate our SILAC approach for proteome quantification results, we re-purified the detergent-insoluble proteome (starting from frozen tissue) from the same control and FTLD cases pooled for proteomics. Immunoblot analysis using commercially available antibodies for CNDP2, TPI1, RHOGDI-alpha, cofilin-1 and cathepsin D showed strong immunoreactivity in all FTLD cases (Fig. 5C and Table 1), however, there was more heterogeneity in the cases which were used for the control pool in the brain/SILAC proteomics experiment. For example, in control case 5, the signal intensities for TPI1, cofilin-1 and RHOGDI-alpha were equivalent to FTLD cases, but were significantly weaker in other control cases. In comparison, CNDP2, which had the greatest log₂ difference (average normalized log₂ =2.47 ±0.48) was barely detected in any of the control insoluble samples. It should be noted that while SILAC comparison identified significant changes in cathepsin D and RHOGDI-alpha which we validated here, these proteins were not considered significantly enriched by SC. Thus, the SILAC internal standard aided in the accurate quantification of these low abundance proteins in the FTLD detergent-insoluble proteome. In contrast to the absolute differences we saw for the above proteins we did not observe significant enrichment of TDP-43 (averaged normalized log₂ = −0.03 ±0.02) in the pooled FTLD samples by proteomics. It is important to note that native unmodified TDP-43 (~43 kDa) is intrinsically detergent-insoluble in control cases as previously demonstrated, whereas the biochemical signature of TDP-43 in FTLD is comprised of lower stoichiometry species (relative to the native 43 kDa species), which are cleaved, phosphorylated or resolve at high molecular weight [3, 49] (Supplemental Figure 4). This largely explains why we did not observe TDP-43 peptides outside of the gel region for the molecular weight range containing the vast majority of TDP-43. Only one TDP-43 peptide was sequenced in a MW region above 60 kDa and none below 25 kDa (see Supplemental Tables S5-8). Thus, lower stoichiometry TDP-43 species (i.e. ubiquitinated or cleaved) in FTLD frontal cortex were below the limit of detection in this experiment. Future studies may require enhanced separation of the insoluble protein extracts by SDS-PAGE and analysis of an increased number of gel fractions, which will reduce the protein complexity prior to in-gel digestion and LC-MS/MS. Those proteins that we did find more enriched represent a class of potential disease makers defined by their detergent insolubility even more than TDP-43, and may have important roles in the processes underlying pathogenesis.

Biochemical and immunohistochemical analysis of PSF in control and FTLD frontal cortex

Two RNA binding proteins, TDP-43 and FUS, are signature proteins of FTLD and ALS [5]. Of the 21 proteins enriched in FTLD by proteomics, one RNA binding protein found was PSF, a nuclear protein and a core component of paraspeckles, ribonucleoprotein bodies with reported roles in transcription and splicing [50]. Similar to TDP-43 and FUS, PSF contains two RNA recognition motifs (RRM) and a low complexity N-terminal glutamine and proline-rich region (Supplemental Figure 5). We examined PSF protein level and localization in FTLD. Native PSF (707 amino acids) is highly basic, and runs anomalously on SDS-PAGE at approximately 100 kDa [51]. Densitometry analysis of the native species in a panel of 20 detergent-insoluble fractions, including an additional number of control and FTLD cases not used for discovery proteomics, shows a significant enrichment in FTLD (p<0.05) (Fig. 6A and B). Only certain FTLD cases showed enrichment, which may represent a unique subtype or variability in regional sampling of the FTLD brain tissue. No correlation between TDP-43 levels and PSF levels were observed in this analysis of detergent-insoluble fraction (data not shown). Smaller PSF species (~25–27 kDa) were also observed in six FTLD cases and one control. The immunoreactivity of the lower molecular weight (MW) species correlated with the intensity of native full-length PSF. Unique PSF peptides were not indentified in lower MW regions (bands C-E), indicating that truncated PSF species were below the limit of detection in this experiment.

In healthy individuals, both TDP-43 and FUS reside in the nucleus and become sequestered in the cytoplasm in disease [5]. To assess PSF subcellular localization in frontal cortex tissue, we performed immunohistochemistry in FTLD and control cases. PSF protein was predominantly nuclear (Fig. 6C), which is consistent with previous reports in mouse brain [52]. Fluorescent labeling of PSF and phosphorylated TDP-43 (pTDP-43) was performed to determine if PSF displays altered distribution in neurons concurrent with cytoplasmic TDP-43 inclusions occurring in FTLD brain (Fig. 6D, upper panel). In layer II neurons of the frontal cortex, PSF (red) was exclusively localized to the nucleus and did not overlap with cytoplasmic pTDP-43 inclusions (green). Consistently, in hippocampal neurons in the dentate gyrus, colocalization between pTDP43 and PSF was not observed (Fig. 6D, lower panel). Importantly, compared to controls, three FTLD cases (5, 7 and 10), displayed a population of cells with profoundly abnormal cytoplasmic PSF distribution including distal processes at the interface of gray and white matter that were morphologically consistent with reactive glia. Although aberrant cytoplasmic PSF was most dramatic in this population of non-neuronal cells, we observed enhanced PSF immunoreactivity in a limited number of neuronal nuclei. Therefore, it remains possible that the increased levels of detergent-insoluble PSF identified by proteomics and validated by immunoblot originate from multiple cell-types, including neuronal and also non-neuronal populations. PSF immunoreactivity was abolished after pre-absorption with a molar excess of immunizing peptide, demonstrating the specificity of PSF antibody (Supplemental Figure 6). Since PSF isoforms that lack the C-terminal nuclear localization sequence (NLS; comprised of amino acids 703–707) due to alternative splicing or proteolytic cleavage display nearly exclusive cytoplasmic localization [41, 51], it is possible that cytoplasmic PSF observed in FTLD could similarly lack the C-terminal NLS. However, the rabbit polyclonal PSF antibody used in this study was raised against a C-terminus epitope (amino acids 650–707). To further characterize this antibody and to determine if the PSF antibody used in this study can detect PSF isoforms lacking the C-terminal NLS, recombinant green fluorescent protein (GFP) tagged, wild-type PSF-WT and PSF-ΔNLS (C-terminal NLS is deleted) were examined in HEK293 cells by immunocytochemistry and immunoblotting (Supplemental Figure 7). As expected, PSF-WT had primarily punctate nuclear localization, whereas PSF-ΔNLS was exclusive to the cytoplasm and in some cells formed aggregates, mimicking the aberrant PSF localization observed in FTLD [41]. Importantly, the polyclonal PSF antibody detected both PSF-ΔNLS and WT-PSF by immunoblotting. These results are consistent with the possibility that the cytoplasmic PSF observed in FTLD may be an isoform of PSF lacking the extreme C-terminal NLS. Alternatively, phosphorylation of full-length PSF by BReast tumor Kinase (BRK) has also been shown to promote PSF nuclear to cytoplasmic translocation [53]. Thus, there remain a number of putative causes for why cytoplasmic PSF is observed in FTLD.

PSF is expressed in neurons, astrocytes and oligodendrocytes

Since it was unclear what non-neuronal cell-type harbored cytoplasmic PSF in brain, we sought to profile PSF protein levels by immunoblotting PSF from total cell lysate from primary cultured rat neurons, astrocytes, oligodendrocytes and microglia (Fig. 7A). Consistent with previous reports, we detected full-length PSF protein in neurons and oligodendrocytes [52], and additionally, PSF was detected in rat and human oligodendrocyte cell lines (CG4 and HOG, respectively) (Fig. 7B). However, contrary to previous findings [52], we observed abundant PSF immunoreactivity in GFAP-expressing astrocytes, with low levels of PSF also detected in primary microglia cells (Fig.7A). Protein levels for polypyrimidine tract binding protein (PTB) and TDP-43 were also analyzed by immunoblot in primary cultured brain cells. PTB, an hnRNP with roles in splicing, was previously identified in non-neuronal cells and characterized to complex with PSF [51]. In contrast to PSF, TDP-43 showed expression in all cell types including microglia, whereas PTB immunoreactivity was only observed in astrocytes and oligodendrocytes. Together, these data indicate that structurally homologous RNA-binding proteins display cell-specific patterns of expression and that astrocytes and oligodendrocytes are the major non-neuronal cell types that express detectable levels of PSF.

Oligodendrocytes harbor cytoplasmic PSF in FTLD

To determine if astrocytes or oligodendrocytes display the cytoplasmic and branched process-specific PSF in FTLD, we performed double-labeled immunofluorescent studies with antibodies against PSF and the astrocyte marker GFAP or oligodendrocyte marker Olig2. In astrocytes, PSF was found predominantly in the nucleus (Fig. 8), and only a small percentage (~11%; 5/44 total cells captured in our analysis) displayed cytoplasmic PSF. These observations indicate that astrocytes are not the major glial cell type harboring cytoplasmic PSF in FTLD. Olig2 is normally located in the nucleus of some resting glial progenitors and the entire lineage of oligodendrocytes, but is not expressed in mature astrocytes [54]. In control brain, Olig2 and PSF localize in the nuclei of oligodendrocytes in discrete puncta (Fig. 9A, top panel). Of cells displaying prominent cytoplasmic PSF localization in FTLD cases, approximately 74% (31/42 cells captured in our analysis) co-localized with cytoplasmic Olig2 (Fig. 9A, bottom panel). Moreover, immunohistochemistry for PSF and Olig2 on adjacent sections revealed a morphologically consistent pattern of cytoplasmic subcellular redistribution (Fig. 9B). This analysis indicates that the aberrant cytoplasmic PSF accumulation in FTLD is predominantly observed in oligodendrocytes and can co-occur with cytoplasmic Olig2 in addition to nuclear Olig2 puncta in many of the same cells.

PSF and PTB levels are dynamic in the optic nerve during early myelin development

Oligodendrocytes are a type of glial cell, constituting approximately 50% of all cells around neurons in the frontal cortex and produce myelin in the white matter, which insulates and enhances the speed by which action potentials are propagated on axons [55]. Severe white matter changes often ensue after acute neurological damage (e.g. stroke), but are also a common feature in chronic progressive neurodegenerative diseases including FTLD [56–58]. Indeed, we noted that myelin basic protein (MBP) and myelin associated glycoprotein (MAG), observed in the brain specific proteome and quantified by SC, were significantly decreased (p<0.01) in FTLD compared to control (Supplementary Table 4). Currently, the function of PSF in oligodendrocytes during normal myelin development is unclear. To assess whether PSF levels change during myelin development, we performed immunoblots for PSF in the mouse optic nerve (Fig. 7C). At birth, the entire optic nerve is unmyelinated and as mice age, the axons in the optic nerve become sheathed by the myelin produced by oligodendrocytes [59]. We observed a concomitant reduction of PSF and PTB levels after postnatal day 23 (p23) and more dramatically at p38, which correlated with an increase in 2′,3′-cyclic nucleotide 3′ phosphodiesterase (CNP); a marker for mature myelin-producing oligodendrocytes. In contrast, TDP-43 levels remain largely unchanged, with only a slight reduction at p38 suggesting that both PSF and PTB act in oligodendrocytes and show a developmentally programmed decline during early myelination.