MS-screen: PCP-SILAC increases the confidence in identifying novel centrosomal proteins

We have previously characterized the protein composition of the human centrosome by using protein correlation profiling (PCP) (Andersen et al, 2003). In this approach, proteins identified by MS are profiled from peptide intensity signals in several gradient centrifugation fractions and distinguished as genuine components when matching a consensus profile determined for known organellar proteins. The principle idea of this method is powerful to sort out unrelated proteins, but the inaccuracy of label-free protein quantitation diminishes its performance; in particular for proteins identified by a few peptides. Thus, to further advance our ability to classify organelle proteins, we aimed at increasing the accuracy of protein quantitation in PCP by introducing stable isotope labelling by amino acids in cell culture (SILAC) (Figure 2A). This was achieved by generating an unlabelled matching internal standard that could be mixed with the corresponding fractions prepared from one or two differentially isotope-labelled cell populations. The method, termed PCP-SILAC, has features distinct from related strategies based on chemical isotope labelling by ICAT (Dunkley et al, 2004; Sadowski et al, 2006), iTRAQ (Borner et al, 2006; Yan et al, 2008), and strategies based on subtractive proteomics (Yates et al, 2005).

Open in a separate window

Figure 2

Identification of centrosomal proteins by PCP-SILAC. (A) Schematic outline of the PCP-SILAC method used to distinguish centrosomal proteins from a background of co-purifying proteins. Centrosomes were isolated by sucrose gradient centrifugation from isotope-labelled and unlabelled cells. The six centrosome-containing fractions collected from the unlabelled cells were pooled to generate an internal standard, which was distributed into the six corresponding fractions collected from the labelled cells before processing these samples for MS analysis. (B) The enrichment of proteins relative to the internal standard is illustrated by the mass spectra of a single peptide (DFLQETVDEK) from the centrosomal protein CEP135 in fractions 1–6 where the peptide isotope clusters are marked by a triangle for signals representing the unlabelled internal standard (light isotope-labelled peptide) and by an asterisk for signals representing the sample in each fraction (heavy isotope-labelled peptide). (C) The enrichment profile of CEP135 was calculated from the isotope ratios shown in (B). (D) Profiles of 32 known centrosomal proteins and the resulting average consensus centrosomal profile. (E, F) Profiles of the DFLQETVDEK peptide from CEP135 and the consensus set of centrosomal proteins were determined from an independent experiment using only four fractions. The 32 proteins co-eluting in both experiments are included in Supplementary Tables S1 and S2).

In practice, two centrosome preparations were isolated in parallel from asynchronously growing human cells cultured in medium containing either normal lysine (Lys0) or ¹³C₆¹⁵N₂ isotope-labelled lysine (Lys8). Fractions collected after the final sucrose gradient centrifugation were tested for the presence of centrosomal proteins by MS analysis of peptides derived from in-solution digests of aliquots taken from each fraction (data not shown). The analysis identified six fractions with detectable levels of centrosomal proteins. These fractions collected from Lys0-labelled cells were mixed to generate a common internal standard for peptide isotope ratio determination. Aliquots of the internal standard were distributed into the corresponding six centrosome-containing fractions collected from Lys8-labelled cells. Proteins in these six samples were then separated by one-dimensional gel electrophoresis, in-gel digested with endoproteinase Lys-C, and the resulting peptides analysed by LC-MS (see Materials and methods).

Mass spectra of the peptide DFLQETVDEK derived from the centrosomal protein CEP135 displayed a large analyte to internal standard ratio (Lys8/Lys0) for the peak centrosomal fraction as compared with spectra of the corresponding peptide in the other fractions (Figure 2B, panel 2). The six isotope ratios allowed us to calculate a protein enrichment profile as the median of the Lys8/Lys0 isotope ratio for all lysine-containing peptides identified for CEP135 in each of the six samples (Figure 2C). Protein enrichment profiles were then calculated for all proteins. A group of 32 known centrosomal proteins were selected for inter-experiment comparison and for determination of a consensus profile for organelle classification (Supplementary Table S1). Profiles of the centrosomal proteins closely followed the CEP135 profile and demonstrate that accurate enrichment profiles can be obtained by PCP-SILAC (Figure 2D). A second independent PCP-SILAC experiment with four fractions and inverted isotope labelling demonstrated that the method can be performed with a variable number of fractions and reproducibly identify centrosomal proteins with a narrow distribution of profiles (Figure 2E and F; Supplementary Table S2).

In a third experiment, we explored the ability of PCP-SILAC to profile simultaneously the enrichment of proteins in two independent centrosome preparations using a third preparation as a common internal standard. The correlated distributions between two different gradients were expected to further increase the confidence in organelle classification. To this end, three cell populations were labelled with different isotopes. The centrosome-containing fractions prepared from unlabelled cells were mixed and used as the common internal standard for the corresponding fractions prepared from each of the two labelled cell populations (Supplementary Figure S1A). The set of 32 known centrosomal proteins were represented by a total of 4661 peptide ratios in the peak fraction and resulted in a narrow distribution of profiles in both experiments (Figure 3A and B). The consensus profiles derived from these data were compared with the profiles of proteasomal and ribosomal subunits, representing co-purifying contaminants residing in structures of different sizes. The profiles were clearly separate from the consensus profiles and displayed consistently altered fractionation behaviour for all proteins in these structures (Figure 3C–F). A goodness of fit was determined as the Mahalanobis distance from the centrosomal consensus profiles, which take into account the variance and the covariance of the measured ratios. Distance scores <9 in both experiments were observed for 110 of the 116 (95%) known centrosomal proteins identified with complete profiles in both experiments, whereas all identified subunits of the proteasome and the ribosome resulted in larger values (>15) (Figure 3G; Supplementary Table S4). Moreover, only three apparent false positive proteins (ALDOA, ALDOC, and ADSL) had distance scores <9. These data demonstrate that the double PCP-SILAC experiment has indeed the ability to distinguish organelle proteins from a background of unrelated proteins with a high degree of confidence on the basis of correlated profile distributions. Data derived from this experiment also demonstrated a clear gain in the sensitivity and specificity of the PCP-SILAC method to distinguish true centrosomal proteins from a large background of unrelated proteins as compared with its label-free version (Supplementary Figures S3 and S4). Importantly, the relative enrichment profile of 1318 proteins quantified in the 2 × 5 fractions revealed a group of 150 proteins that fulfilled the stringent criteria of distance scores <9 in both experiments (Figure 3G). With few exceptions, this list comprises the majority of known centrosomal proteins (110 proteins) including the 23 novel proteins reported in our previous study (Andersen et al, 2003) (Table I; Supplementary Tables S3 and S4).

Open in a separate window

Figure 3

Identification of centrosomal proteins by the double PCP-SILAC experiment. (A, B) Centrosomes were isolated by sucrose gradient centrifugation from three different isotope-labelled cell populations to profile the elution of proteins in two separate preparations simultaneously in a single experiment using one of the preparations as an internal standard (see outline of the double PCP-SILAC experiment in Supplementary Figure S1). The profiles for 32 known centrosomal proteins follow a narrow enrichment profile in both preparations and demonstrate that these proteins co-elute. The shape of the profiles is not critical for organelle classification but reflects a shift in the elution of proteins between the two experiments. (C–F) The profiles of proteasomal and ribosomal subunits obtained from the same data set are distinct from the centrosomal consensus profiles. (G) An organelle classification score was calculated as the Mahalanobis distance between the centrosomal consensus profile and all other proteins with a complete enrichment profile in the double PCP-SILAC experiments 3A and 3B. Known centrosomal proteins and likely candidates clustered in a region with distance scores <9 as compared with, for example, proteasomal and ribosomal subunits with high distance scores.

Image analysis of cells stably expressing GFP-fusion proteins or stained with antibodies confirms centrosome localization for candidate proteins identified by PCP-SILAC

To confirm the in vivo subcellular localization of the identified MS-candidates at any stage of the cell cycle, we stably expressed N- or C-terminally tagged green fluorescent fusion proteins at their endogenous levels in HeLa Kyoto cells using BAC TransgeneOmics (Poser et al, 2008). The resulting cell pools were immunostained with anti α- or γ-tubulin antibodies to visualize centrosomes and anti-GFP antibody to enhance the fluorescence signal of the tagged proteins. In vivo localization to centrosomes and spindles were observed for 14 of 27 tested candidate proteins (Figure 4, Supplementary Figure S5; Table I, Supplementary Table S4). MS-candidate proteins were also confirmed by immunofluorescence microscopy in U-2 OS cells using HPA antibodies. Localization to centrosomes were observed for 11 of 16 tested candidate proteins (MPHOSPH9, C6orf204, SLAIN2, CCDC46, Albatross, C14orf145^*, CCDC45^*, MIB1^*, KIAA0753^*, CCDC21^*, and GARNL4). Candidates marked by an asterisk were also confirmed in HeLa cells stably expressing the corresponding GFP-tagged fusion protein. Images are available at http://www.cebi.sdu.dk/CepDB. In general, we observed a good correlation between the GFP and antibody staining, although the signals for the GFP-tagged proteins were often weaker and in some cases too weak (e.g., MPHOSPH9, C6orf204, and SLAIN2) to confirm centrosome association. Staining of structures around centrosomes similar to centrosome satellites was observed for CCDC123^*, PRKACB, and CCDC14A.

Open in a separate window

Figure 4

Protein candidates identified by the MS-screen localize to the centrosome. (A) Co-staining with the centrosomal marker protein γ-tubulin (Cy3) support centrosome (two dots) or centriole (four dots) localization for the indicated candidates (green). Additional candidates validated as novel centrosomal proteins are listed in Table I and Supplementary Tables S4 and S5. (B) GFP–C3orf34 co-localize with γ-tubulin (Cy3) at all major stages of the cell cycle in HeLa cells. DNA was stained with DAPI, yellow indicates coincidence of green and red signals. Bars, 1 μm (insert) and 5 μm.

To identify centriole-associated proteins, we counted the number of structures stained in the centrosome area. In most cases, two dots co-localizing with the pericentriolar marker protein γ-tubulin were observed (Figure 4A; Supplementary Figure S5). GFP–KIAA0562, however, was resolved in four dots suggesting that this protein localize to centriole pairs in duplicated centrosomes comparable to, for example, CEP97 (Figure 4A). MPHOSPH9 stained two–six dots in non-separated centrosome pairs. This unusual staining was interpreted as proximal and distal staining of centrioles when comparing with the staining of γ-tubulin (Figure 4A; Figure 6D). The antibody specificity was supported by reduced staining in cells depleted for MPHOSPH9 by esiRNA (data not shown).

Comparison of the staining patterns of centrosomes between interphase and mitotic stages revealed that the majority of candidate proteins localized differentially to centrosomes, spindles, and midbody during the cell cycle (Supplementary Table S4; Supplementary Figure S5). This is illustrated by images of C3orf34 at different stages of the cell cycle. C3orf34 localized asymmetrically to centrosome/centriole pairs in early interphase, to spindle poles during mitosis, and to distinct foci oriented towards the midbody at telophase (Figure 4B).

HPA-screen: evaluation of confocal images of three different cell lines stained with HPA antibodies identifies additional centrosome candidate proteins

Centrosomes are plurifunctional organelles with specialized roles in various cell types. To characterize the centrosome proteome representing more than a single cell type, we performed a complementary antibody-based screen in U-251 MG glioblastoma, A-431 epidermoid carcinoma, and U-2 OS osteosarcoma cells. Proteins were stained with HPA antibodies covering >4000 genes and the resulting immunofluorescence images were initially evaluated for centrosome localization using α-tubulin as a marker for microtubules. The 4000 genes comprised 46 genes that were already identified as known or candidate centrosomal proteins by the MS-screen. This subset was further evaluated by γ-tubulin co-localization in U-2 OS cells. We found that 14 of these antibodies validated the novel MS-candidates described above and that 26 of 30 antibodies representing known centrosomal proteins clearly supported centrosome association (Supplementary Table S5). These experiments ascertained the quality and value of the HPA antibodies for detecting known and novel centrosomal proteins.

A combination of automated and manual annotation of images acquired for the remaining HPA antibodies indicated centrosome staining for an additional 113 proteins in at least one of the three cell lines. The immunofluorescence images were scored in comparison to a negative control. Any staining stronger than the control was assigned one or multiple location as indicated in Supplementary Table S5. Centrosome association were assigned when at least two cells displayed the characteristic two-dot staining pattern of disengaged centrioles or duplicated centrosomes and/or when the protein staining clearly co-localized with α-tubulin at microtubule organising centres. Centrosome localization varied from fine two-dot structures to more diffuse structures or distinct satellites in the surrounding area of centrosomes (Figure 5). We selected a subset of 21 of these HPA candidates representing primarily the former group and tested for γ-tubulin co-localization in U-2 OS cells. Ten candidates co-localized with γ-tubulin (PRICKLE3, PCDH19, TNFAIP3, PCDHB15, OAZ1, SERPINB6, BTN3A3, CHST15, KIAA1107, and NAV1). An additional four candidates (DDHD2, TECTA, LGI2, and TWSG1) stained structures in the vicinity of centrosomes. Half of the 100 candidate proteins were initially observed in only one of the three cell lines. FRMD5, for example, displayed only intense centrosomal straining in U-251 MG cells but not in U-2 OS cells where it clearly stained the midbody (Supplementary Figure S7A). OAZ1 stained satellites concentrated around centrosomes in A-431 cells. A more defined localization to basal bodies was apparent from the staining of motile cilia in ciliated tissues and primary cilia and basal body in hTERT-RPE1 cells (Figure 5). These experiments demonstrate that the HPA-antibody screen is complementary to the MS-screen and suggest that genuine centrosomal proteins worthy of further study are likely to be found from the list of HPA candidates (Supplementary Table S5; http://www.proteinatlas.org). Although the HPA antibodies were epitope purified and evaluated for specificity and assigned reliability score (see Materials and methods), we cannot exclude false assignments due to fortuitous cross-reaction (Nigg et al, 1982). Hence, caution needs to be exerted when interpreting the results of the antibody screen and additional experiments are required to confirm the HPA candidates.

Open in a separate window

Figure 5

HPA-antibody screen in three different cell lines identify additional centrosomal proteins. Antibody staining of OAZ1 in U-251 MG, A-431, and U-2 OS cells suggests centrosome localization. Staining of OAZ1 in serum starved hTERT-RPE1 cells and nasopharynx tissue support basal body and cilia (primary and motile) localization. The centrosomal marker protein γ-tubulin (or α-tubulin) was stained with Cy3 and DNA with DAPI, yellow indicates coincidence of green and red signals. Bars, 5 μm. The immunohistochemical staining (brown–black) was counterstained with haematoxylin (blue colouring of both cells and extracellular material) to enable visualization of microscopic features.

Image analysis of cells at different stages of cell-cycle progression identifies proteins localizing selectively to the mother centriole

Ultrastructural analysis of centrosomes and cilia has revealed proteins localizing to functionally relevant substructures such as the distal and subdistal appendages of the mature mother centriole involved in anchoring of microtubule and docking of basal bodies at the plasma membrane (Marshall, 2008; Debec et al, 2010). These proteins are often asymmetrically associated with the centrosome during the cell cycle. To identify proteins with this behaviour, we examined images with two closely spaced dots marked by γ-tubulin. The two dots likely represent already duplicated centrosomes or individual centrioles disengaged for procentriole formation where the distance between the mother and daughter centriole is sufficiently large to be resolved by light microscopy. Differential staining of the two dots at the G1–S–G2 phases was observed for C14orf145, BTN3A3, PRICKLE3, Albatross, C3orf34, ODF2, and CNTROB. Thus, these proteins might selectively associate with the mother or the daughter centriole. C14orf145 and Albatross were shown to localize to the distal end of the mother centriole in hTERT-RPE1 cells serum starved for 48 h to induce primary cilium formation from the mother centriole (Figure 6A and B). In contrast, PRICKLE3 localized to a single centrosome throughout interphase, even after centrosome splitting. Moreover, PRICKLE3 was observed at the spindle poles during mitosis and stained both the mother and daughter centrioles in ciliated hTERT-RPE1 cells (Figure 6C). C3orf34-GFP co-localized with ODF2 at the mother centriole (Figure 6D) whereas BTN3A3 was shown to localize to the daughter centriole in ciliated hTERT-RPE1 cells (Supplementary Figure S7B). Additional experiments are, however, required to confirm this finding because a second HPA antibody for BTN3A3 revealed non-centrosomal staining. Interestingly, MPHOSPH9 stained three dots in ciliated hTERT-RPE1 cells, suggesting that it localize to the proximal and distal end the daughter centriole, the proximal end of the mother centriole, and no longer to the distal end of the mother centriole forming the axoneme structure (Figure 6D, panel 3).

Open in a separate window

Figure 6

Proteins identified by the MS- and the HPA-screen display asymmetric centrosome localization. (A) C14orf145, (B) Albatross, and (C) PRICKLE3 stain a single centrosome/centriole at the G1/S-phase of the cell cycle in U-2 OS cells as compared with the centrosomal marker protein γ-tubulin (Cy3). To distinguish between mother and daughter centriole association, the axoneme-extended mother centriole were stained with anti-acetylated tubulin (Cy3) in ciliated hTERT-RPE1 cells. This indicates that C14orf145 and Albatross localize to the mother centriole. (D) Antibody staining of GFP–C3orf34 in HeLa cells at the G1/S-phase and MPHOSPH9 in RPE cells at the G0 phase. Co-staining with the mother centriolar marker protein ODF2 (Cy5) indicates that GFP–C3orf34 associate with the mother centriole. Co-staining with anti-acetylated tubulin (red) in ciliated hTERT-RPE1 cells suggest that MPHOSPH9 (green) localize proximal at the mother centriole and distal and proximal at the daughter centriole. DNA was stained with DAPI, yellow indicates coincidence of green and red signals. Bars, 5 μm. (E, F) Interpretation of the staining patterns of tubulin (red) and asymmetrically localized candidate proteins (green) in duplicated centrosomes and basal bodies.

Rough estimates of the relative protein abundance indicate the subunit stoichiometry of centrosome subcomplexes

Quantitation of proteins by MS provides the opportunity to examine the biochemical properties of these candidates in more detail. To roughly estimate the relative protein abundance of proteins in isolated centrosomes we made use of the peptide intensity signals from the double PCP-SILAC experiment. As the intensity signals between different peptides display large variation we averaged as many signals as possible and calculated the centroid of protein abundance profiles to improve reliability. The resulting data indicated that the abundance of centrosomal proteins spans at least two orders of magnitude (Figure 7A and B; Supplementary Table S3). This is exemplified by PLK4 with a relative abundance of 4% as compared with γ-tubulin with a relative abundance near 100%.

Open in a separate window

Figure 7

Abundance and turnover of centrosomal proteins are shown. (A) Rough estimates of protein abundance calculated from the averaged peptide intensities for known and candidate centrosomal proteins identified by PCP-SILAC (average of two experiments). (B) Relative protein abundance of selected centrosomal proteins. (C) Schematic outline of the pulsed-SILAC experiments to measure the turnover rate of centrosomal proteins. The experiment was performed in the absence and presence of a fully Lys-²H₄-labelled internal standard. (D, E) Mass spectra of pulsed-labelled TUBG1 and AURKA peptides. White and grey circles indicate the old (light isotopes, L) and the newly synthesized pool (heavy isotopes, H) of proteins, respectively; dark circles indicate the internal standard (medium isotopes, M). (F) Turnover rates for known and candidate centrosomal proteins (average of two experiments). Highlighted proteins are the kinases PLK1, PLK4, AURKA, and NEK2, and the γ-TuRC subunits. (G, H) The turnover rates for the subunits of the γ-TuRC and HAUS complexes, respectively, are shown in black. Grey bars indicate the normalized sum of the old and newly synthesized pool of proteins (L/M+H/M), which provide a control for the measured ratio.

The roughly estimated relative protein abundances were consistent with the stoichiometry reported for the γ-tubulin small complex (γ-TuSC), which is composed of two molecules of TUBG1 associated with one molecule each of TUBGCP2 and TUBGCP3 (Figure 7B). Multiple copies of the γ-TuSC proteins associate with additional proteins, including TUBGCP4, TUBGCP5, TUBGCP6, and NEDD1 to form the γ-TuRC. The stoichiometry of this complex is less well established. Our data indicate that NEDD1 is equally abundant to TUBGCP2 and TUBGCP3, whereas TUBGCP4, TUBGCP5, and TUBGCP6 are substoichiometric (Figure 7B).

Isolation of centrosomes

Centrosomes were isolated as described by Moudjou and Bornens (1994), see supplemental methods. Centrosome-containing fractions (0.4 or 0.5 ml) were identified by LC-MS of peptide mixtures derived from in-solution digests of 25 μl aliquots of each fraction (see Supplementary data).

PCP-SILAC

The centrosome-containing fractions from unlabelled cells (Lys0) were mixed to generate a matching internal standard. The combined fractions were diluted with 10 mM Pipes buffer (pH 7.2) and aliquots were added to fractions containing centrosomes from SILAC-labelled cells (Lys8). Centrosomes from the resulting samples were pelleted by centrifugation at 16 000 g for 15 min. In an inverted labelling experiment, centrosomes from Lys8-labelled cells were mixed as the matching internal standard and divided into four fractions containing centrosomes from unlabelled cells. In a separate double PCP-SILAC experiment, centrosomes from unlabelled cells were mixed as the matching internal standard and divided into the 2 × 5 centrosome-containing fractions isolated from Lys4+Arg6-labelled cells and from Lys8+Arg10-labelled cells. Centrosomes from the final samples were pelleted by centrifugation by centrifugation at 16 000 g for 15 min. The centrosomal proteins were separated by SDS–PAGE (see Supplementary Figure S1A) and in-gel digested with endoprotease Lys-C or trypsin. The resulting peptides were extracted and desalted for LC-MS analysis (see Supplementary data).

MS and peptide identification

MS analysis was performed by LC-MS using an Agilent HP1100 system and a linear ion-trap Fourier-transform ion-cyclotron resonance (LTQ-FT-ICR) or an LTQ-Orbitrap mass spectrometer (Thermo Fisher) (see Supplementary data). Peak lists for protein database searches (Human IPI version 3.52) were extracted from the resulting data using MSQuant, an in-house developed open source application (http://msquant.sourceforge.net/), or MAXQuant (version 1.0.12.16) (Cox and Mann, 2008). Both programs were also used to calculate peptide enrichment ratio and to evaluate the certainty in peptide identification and quantitation based on Mascot score, false discovery rates, MS³ scoring (Olsen and Mann, 2004), or manual inspection. Initially all peptides with a Mascot score of at least 15 were quantified automatically to include as many peptides in all fractions as possible. Proteins included in Table I are based on proteins identified by at least two unique peptides and a false discovery rate of 0.01 derived by decoy database searching (Cox and Mann, 2008). Tools for direct evaluation of PCP-SILAC experiments were implemented in MSQuant, including the display of protein enrichment profiles, calculation of consensus profiles of selected proteins, and a score for organelle classification (Supplementary Figure S2).

Statistical and bioinformatic analysis

The relative enrichment of quantifiable peptides in each of the sucrose gradient fractions was calculated from the peak area of the ‘light/heavy’ isotope ratio for each single scan mass spectrum. For each experiment, the median of log2 transformed peptide ratios was computed for each protein in each fraction. Ratios of a ‘consensus set’ of 32 known centrosomal proteins, against which all other proteins were compared, were used to calculate an organelle classification score (see Supplementary data). The relative abundance of protein was estimated from the averaged intensities of all peptide m/z signals associated with each protein in each fraction and calculated as the centroid of the resulting abundance profiles. The percentage of pulsed isotope labelling was calculated as (1/((1/ratio)+1)) × 100. ProteinCenter (Proxeon A/S) was used for clustering of proteins identified across independent experiments and for comparison with published data sets of proteins from centrosome-related structures with 98–95% sequence homology.

Cloning, tagging, and expression of fluorescent proteins

Cell pools with N- or C-terminally tagged enhanced green fluorescent protein (EGFP) were generated using BAC TransgeneOmics as previously described (Poser et al, 2008). For a smaller group of selected candidates, cDNAs were obtained from the Kazusa DNA Research Institute, Japan and cloned N- or C-terminally to EGFP by using the Gateway cloning system (Invitrogen). U-2 OS cells were cultured, transfected with the DNA constructs using calcium phosphate precipitates, and grown in selection medium containing G418 (1600 μg/ml) to establish cell lines stably expressing EGFP-tagged proteins.

Immunofluorescence microscopy

HeLa Kyoto cells stably expressing EGFP-tagged proteins were cultured on glass coverslips and fixed in cold methanol (−20°C) for at least 5 min and rehydrated in phosphate buffered saline (PBS) before blocking in 1% horse serum, 0.1% Tween-20 in PBS for 10 min. Cells were stained with anti α-tubulin or mouse anti γ-tubulin (GTU-88, Sigma) and goat anti-GFP antibody for 1.5 h followed by the secondary donkey-anti-mouse Cy3 (Jackson ImmunoResearch) and chicken-anti-goat Alexa 488 antibodies (Invitrogen) in combination with DAPI for 1 h. hTERT-RPE1 cells were serum starved for 48 h before fixation and staining with mouse anti-acetylated tubulin (clone 6-11B-1, Sigma) together with HPA antibodies at a final concentration of 2 μg/ml. In the HPA project, immunofluorescence microscopy was systematically used to determine protein subcellular location in U-251 MG, A-431, and U-2 OS cells using in-house produced antibodies, as previously described (Barbe et al, 2008). The antibodies were epitope purified and have been evaluated for specificity and assigned a reliability score (supportive, uncertain, and not-supportive) in protein arrays, western blot, immunohistochemistry, and immunofluorescence microscopy. The validation data as well as the antigen sequence are available in the HPA. See http://www.proteinatlas.org/about/assays+annotation and http://www.proteinatlas.org/about/quality+scoring for more information. Selected HPA candidates were also co-stained with anti-γ-tubulin antibodies in U-2 OS cells. Coverslips were mounted on glass slides and imaged using a Zeiss Axiovert 200M laser scanning confocal microscope LSM510 or a Zeiss CellObserve and an Apochromat × 63 1.4 n.a. oil-immersion objective.

We thank members of the Center for Experimental BioInformatics (CEBI), the Hyman, the Uhlen, and the Nigg group for fruitful collaboration, discussions, and critical reading of the manuscript; Rikke Jakobsen, Mogens M Nielsen, and Peter Mortensen for excellent technical assistance; Proxeon A/S for the use of ProteinCenter; Mark Payne, Technical University of Denmark for statistical data analysis; and Mikael Le Clech for initial bioinformatics exploration. The research leading to these results has received funding from The Lundbeck Foundation, The Danish Agency for Science, Technology and Innovation, and the European Commission's 7th Framework Programme (grant agreement HEALTH-F4-2008-201648/PROSPECTS. AAH acknowledges funding by the Max Planck Society, and the BMBF grant DiGtoP (01GS0859). EAN acknowledges support from the Max-Planck society; EL, MS, and MU acknowledge the staff working in the HPA project and the Knut and Alice Wallenberg foundation for funding.

Author contributions: LJ, KV, and JSA designed and performed the mass spectrometry and microscopy experiments and wrote the manuscript. MS, EL, and MU generated antibodies and performed the HPA-screen. IP, YT, and AAH generated cell pools expressing GFP-tagged proteins and performed microscopy experiments. LGF and MB performed bioinformatics analysis. JW and EAN contributed to the experimental design and data interpretation.