Interaction of Human HSP22 (HSPB8) with Other Small Heat Shock Proteins*

Mammalian small heat shock proteins (sHSP) are abundant in muscles and are implicated in both muscle function and myopathies. Recently a new sHSP, HSP22 (HSPB8, H11), was identified in the human heart by its interaction with HSP27 (HSPB1). Using phylogenetic analysis we show that HSP22 is a true member of the sHSP superfamily. sHSPs interact with each other and form homo- and hetero-oligomeric complexes. The function of these complexes is poorly understood. Using gel filtration HPLC, the yeast two-hybrid method, immunoprecipitation, cross-linking, and fluorescence resonance energy transfer microscopy, we report that (i) HSP22 forms high molecular mass complexes in the heart, (ii) HSP22 interacts with itself, cvHSP (HSPB7), MKBP (HSPB2) and HSP27, and (iii) HSP22 has two binding domains (N- and C-terminal) that are specific for different binding partners. HSP22 homo-dimers are formed through N-N and N-C interactions, and HSP22-cvHSP hetero-dimers through C-C interaction. HSP22-MKBP and HSP22-HSP27 hetero-dimers involve the N and C termini of HSP22 and HSP27, respectively, but appear to require full-length protein as a binding partner.

In a previous publication we identified mammalian HSP22 (other names: HSPB8, H11, E2IG1), a protein similar to known small heat shock proteins (sHSP), 1 by a two-hybrid (TH) screen using HSP27 (HSPB1, HSP25), the classic heat-inducible sHSP, as "bait," and a heart cDNA library (1). While in this and other reports HSP22 was characterized as an sHSP (2,3), in some reports the newly described protein was classified as a protein kinase with similarity to the Herpes simplex protein kinase ICP10 (4 -6). HSP22 occurs preferentially in striated and smooth muscles, but also in brain (1), estrogen receptorpositive breast cancer cells (2), melanoma cells (4), and keratinocytes (5). In addition to HSP22, also abundant in muscles are other sHSPs including HSP27, myotonic dystrophy kinasebinding protein (MKBP, HSPB2), HSPB3, ␣B-crystallin (␣B-Cry, HSPB5), HSP20 (HSPB6), and cvHSP (HSPB7) (7,8). It is now widely accepted that sHSPs play a major role in muscles, although their precise role is not understood. A point mutation in the ␣B-Cry gene causes desmin-related cardiomyopathy in humans (9), and MKBP binds to and activates the myotonic dystrophy protein kinase, an enzyme that when absent results in myotonic dystrophy (10). Overexpression of HSP22 in heart was recently shown to be associated with hypertrophy (6). Also, the specific location of sHSPs in the sarcomers (11) and the protection of myocytes by sHSPs from ischemic stress (12) suggest an important role for these proteins in muscles. Phosphorylation of HSP27 has been implicated in the contraction (13), and phosphorylation of HSP20 in the relaxation (14) of smooth muscle cells.
sHSPs have been shown to form two types of hetero-oligomeric complexes in striated muscles: type I complex consisting of HSP27, ␣B-Cry, and HSP20, and type II complex consisting of MKBP and HSPB3 (8). sHSP complexes are heterogeneous in size and composition. The molecular mass of cellular sHSP complexes varies over a wide range (ϳ50 -1000 kDa) and changes under stress conditions leading usually to smaller complexes (15). The crucial role of these complexes for cell survival under stress conditions has been demonstrated (16). Based on these findings it is believed that the formation of homo-and hetero-oligomeric complexes of sHSPs is essential for their function.
Most available data concerning formation and structure of sHSP complexes are based on studies of ␣A-crystallin, ␣B-Cry, and HSP27. These three sHSPs form dynamic homo-and hetero-oligomeric complexes which may have a micellar structure and rapidly exchange subunits (17)(18)(19). HSP27 dimerizes, and two such dimers form tetramers, which are in equilibrium with larger complexes (20). Human HSP27 is phosphorylated at three sites (Ser 15 , Ser 78 , Ser 82 ) by the protein kinase MAP-KAPK-2 resulting in the disassembly of large oligomeric structures predominantly into tetramers (15,21). Interactions within HSP27 complexes involve at least two sites, one site within the C-terminal ␣-crystallin domain and the other site at the far N terminus (22). The site in the ␣-crystallin domain has been proposed to be involved in dimer formation (23), and these dimers are thought to further multimerize into larger complexes using the N-terminal dimerization site.
If HSP22 belongs to this group of proteins, we hypothesized that it interacts with itself and with other sHSPs which may result in the formation of homo-and hetero-oligomeric complexes, similarly to other sHSPs. In an initial effort to characterize HSP22, we have studied its phylogenetic relationship among Bilateria proteins and confirmed that it is a member of the superfamily of sHSPs. By gel filtration high performance liquid chromatography (HPLC) we show that HSP22 forms high molecular mass complexes in the heart, similarly to other sHSPs. By yeast TH experiments, co-immunoprecipitation (co-IP), cross-linking (CL), and fluorescence resonance energy transfer (FRET) microscopy we show that HSP22 interacts with itself and with the other tested sHSPs. We show also that HSP22 has two binding domains that are specific for individual sHSPs.

EXPERIMENTAL PROCEDURES
Phylogeny of sHSPs-A profile hidden Markov model in HMMR 2.2 g was employed to characterize, search for and align diverse sHSP superfamily members. The non-redundant protein data base (GenBank TM CDS translationsϩPDBϩSwissProtϩPIR) was queried with sHSP-HMM for significant matches (proteins with bit scores of 4 or greater and expectation values of 0.05 or less were retained). The resulting sequences were aligned with the sHSP-HMM. To estimate phylogeny and to approximate the posterior probabilities of the tree, a Bayesian inference approach with Metropolis-coupled Markov chain Monte Carlo, or MC 3 , was used. More details are given elsewhere (24).
Origin of sHSP cDNAs and Plasmid Constructs-All relevant data on origin of sHSP cDNAs, plasmid constructs, used cloning methods and PCR primers are given in Table I. The PCR cycle conditions for cvHSP cDNA amplification from a human heart cDNA library (Clontech) were 50 s 95°C, 50 s 63°C, and 60 s 72°C. For all TH experiments, the vectors pAS2-1 (abbreviated: pAS) and pACT2 (abbreviated: pACT) (Clontech) were used. For eukaryotic expression of Myc-and FLAGtagged sHSPs the vectors pcDNA3.1-myc (Invitrogen) and pFlag-CMV2 (Sigma) were used, and for eukaryotic expression of YFP-and CFP-sHSP fusion proteins the vectors pEYFP and pECFP (Clontech) were used. All plasmid constructs were verified by sequencing. sHSP cDNAs were separated into N-and C-terminal fragments in the central region or beginning of the ␣-crystallin domain as indicated in Fig. 1A.
Two Hybrid Experiments-Small scale sequential transformation of the yeast strain Y190 was performed as described in the manufacturer's instructions (Clontech). Interactions between full-length (F), N-terminal (N), and C-terminal halves (C) of sHSPs were analyzed. Briefly, yeast was first transformed with the constructs 2, 5, or 6 and grown on -Leu medium, or with construct 1 and grown on -Trp medium. In the second step, the yeast was transformed with vectors as specified in the Figs., and plated on -Trp, -Leu, -His selective medium. Y190 has two reporter genes (hisϩ, growth of colonies; galϩ, blue colonies), and two proteins were considered as interacting partners only if both reporter genes were activated. In order to reduce false positive results, every used vector with an insert was tested with an empty partner vector. In some of these controls the hisϩ reporter gene was moderately activated resulting in between 10 and 100 colonies/10-cm Petri dish; however, the galϩ reporter gene was never activated (not shown).
Cell Culture, Transfection, Co-immunoprecipitation, and Cross-linking-293T and COS-7 cells (ATCC) were maintained at 37°C in a 5% CO 2 humidified atmosphere in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum. To prevent detachment of 293T cells, tissue culture plates were treated with poly-L-lysine before seeding the cells. For transient transfection FuGENE 6 (Roche Applied Science) was used according to the manufacturer's protocol. 48 h later, cells were harvested and processed for co-IP, CL, or FRET analysis.
For co-IP, 2 g of a rabbit anti-FLAG polyclonal antibody (Sigma) or mouse anti-Myc monoclonal antibody (Roche Applied Science) were bound to 50 l of protein G-Sepharose (20% slurry, Sigma) by incubating for 2 h at 4°C in 500 l of buffer A (50 mM Tris-HCl, pH 8.0; 1% Triton X-100; 5 mM EDTA; 1 mM EGTA; 1ϫprotease inhibitor mix, Roche Applied Science). Transfected and non-transfected COS-7 cells were rinsed with ice-cold PBS and lysed in buffer A on ice for 30 min. The lysate was centrifuged at 14,000 ϫ g for 10 min at 4°C, and the supernatant was then precleared by incubation with protein G-Sepharose for 2 h at 4°C before it was incubated with the antibody-coated beads overnight at 4°C on a rotating platform. Then the beads were collected by brief centrifugation and washed three times with buffer A. Bound immunocomplexes were released from the beads by a 3-min boiling in 100 l of buffer B (62.5 mM Tris-HCl, pH 6.8; 2% SDS; 10% glycerol; 200 mM dithiothreitol; 0.01% bromphenol blue; 1ϫprotease inhibitor mix). Thereafter, samples were analyzed using SDS-PAGE followed by Western blotting. Negative controls were run with rabbit and mouse preimmune sera instead of anti-FLAG or anti-Myc antibodies (not shown).
For CL, transfected 293T cells grown in 6-well plates were washed with cold PBS. Glutaraldehyde (Sigma) was diluted with PBS (0.0001, 0.0002, 0.0005, 0.001, 0.002% final concentrations) immediately before use, and 1 ml of these solutions was added per well. Cells were incubated for 1 h at room temperature, and then cells were cooled on ice and washed with ice-cold PBS. Cells were lysed with 1 ml of buffer B. The lysates were sonicated for 10 s in order to break down DNA, and then boiled for 5 min. These samples were then analyzed by SDS-PAGE and Western blotting.
Size Exclusion HPLC-0.5 g of the left ventricle of a Rhesus monkey heart was homogenized in 2 ml ice-cold buffer C (50 mM Tris-HCl, pH 7.4; 150 mM NaCl; 5 mM KCl; 1 mM EDTA; 10% glycerol; 1% CHAPS; 1ϫ protease inhibitor mix) using a PowerGen 125 homogenizer. The homogenate was centrifuged at 14,000 ϫ g for 20 min at 4°C, and 100 l of the supernatant was used for HPLC chromatography. Before loading the sample, the column (Protein PAK 300 sw, 0.75 ϫ 30 cm, Waters) was equilibrated with buffer D (50 mM Tris-HCl, pH 7.4; 150 mM NaCl; 5 mM KCl; 1 mM EDTA), and separation of the proteins was at a flow rate of 1 ml/min. Fractions were collected as indicated in Fig. 2. Proteins in all fractions were precipitated overnight by 2 vol of ethanol/1% ␤-mercaptoethanol at Ϫ20°C, and the precipitates were collected by centrifugation. The air-dried pellets were dissolved in 50 l of buffer B, briefly sonicated and boiled for 5 min. These samples were then analyzed for the presence of HSP22, HSP27, and MKBP by SDS-PAGE and Western blotting using specific antibodies. The column resolved proteins with molecular masses ranging from Ͻ10 to Ͼ670 kDa and was calibrated using the molecular mass markers Dextran blue (Ͼ1000 kDa), thyroglobulin (670 kDa), ␥-globulin (158 kDa), ovalbumin (44 kDa), myoglobin (17 kDa), and vitamin B 12 (1.35 kDa) (Fig. 2).
Electrophoretic Methods and Western Blotting-SDS-PAGE and transfer of the separated proteins onto polyvinylidene difluoride membranes (Western blotting) were performed as described (1). For Western blotting, rabbit anti-FLAG antibody (1:5,000) (Sigma), mouse anti-Myc antibody (1:5000) (Roche Applied Science), and sheep anti-HSP22 (1) were used as primary antibodies. As secondary antibodies, goat antirabbit IgG (Fc fragment-specific), goat anti-mouse IgG (Fc fragmentspecific) and rabbit anti-sheep IgG antibodies conjugated to horseradish peroxidase (Jackson) were used at a dilution of 1:10,000. The membranes were processed for detection using the ECL plus reagent (Amersham Biosciences).
Fluorescence Resonance Energy Transfer Measurements-COS-7 cells grown on glass coverslips were transfected with various YFP-sHSP and CFP-sHSP constructs as listed in Table I coding for the corresponding fusion proteins. Two days later, cells were fixed with 4% formaldehyde in PBS, pH 7.4, for 30 min at room temperature. The cells were then washed with PBS, and the coverslips were mounted on slides with Prolong antifade mounting medium (Molecular Probes) and used for microscopy. For FRET microscopy, an Axiovert 135TV microscope equipped with a ϫ40 FLUAR 1.3NA oil immersion objective lens and a CFP/YFP filter set was used (Zeiss). Images were recorded at room temperature using a cooled integrated CCD camera DAGE RT 3000 (DAGE-MTI Inc.). NIH Image software was used to acquire 8-bit gray scale images from a Scion LG3 video capture board and to quantify the image pixel intensities within user selected regions of each cell. The same region was selected for analysis in each of the images obtained from an individual cell. Individual images were exported in TIFF format and image montages created using Adobe Photoshop. No postprocessing of images was conducted after acquisition.
FRET was measured by determining the CFP emission before and after an 8-min photobleaching step of YFP as described (25). The increase or dequenching of CFP emission is a direct measure of the FRET due to interaction. Interaction as measured by FRET is expressed as FRET Factor (FF) in Equation 1.
The symbols are defined as: I c-post , I c-ante , intensity of the CFP-signals in the region of interest after and before photobleaching, respectively; B post , B ante , background signals, outside of cell areas, after and before photobleaching, respectively; CF, correction factor (the ratio of the intensities of the CFP signals after and before photobleaching of an image area with cells which were not photobleached). The CF is used to eliminate fluctuations of the CFP signal during the course of the experiment. Ideally, if two partners do not interact, the FF-value would be 0, and any value above 0 would indicate interaction. However, individually expressed CFP and YFP control proteins are known to interact to a certain extent (26). The FF value of this negative control (FF ϪC ) was determined to be 0.096 Ϯ 0.023 under the applied experimental conditions. Every FF value measured for interacting CFP/YFP-sHSP fusion proteins must be significantly higher than this value. For positive control, a CFP-YFP fusion protein was expressed (27). The measured FF value of 0.384 Ϯ 0.037 is significantly different from FF ϪC (p Ͻ 0.0001) and indicates the range in which FF values for interacting proteins can be expected. For both, the interacting pairs of sHSPs and the controls, the CFP signal intensities before and after photobleaching of 10 randomly selected cells were measured. This allowed determining the significance of the differences of the FF values to the FF ϪC by the Student's t test.

RESULTS
Phylogenetic Relationship of HSP22 among the Bilateria sHSPs-In the light of the divergent roles proposed for HSP22 (see Introduction and "Discussion"), we sought to gain additional information by a phylogenetic analysis. HSP22, like the other human sHSPs, consists of the conserved ␣-crystallin domain, a less conserved N-terminal domain, a variable central region, and variable N-and C-terminal tails (Fig. 1A). Based on the multiple alignment of all 10 human sHSPs, a hidden Markov model (sHSP-HMM) was built that was used to search the entire non-redundant protein data base (24). This sHSP-HMM identified 167 Bilateria sHSP-like proteins, including all known mammalian sHSPs as proteins with significant sequence similarity. The sHSP-HMM was used to align these 167 proteins, and the alignment was used to build a Byesian inference phylogeny. An abbreviated version of this phylogeny illustrating the overall topology of Bilateria sHSPs while detailing the relationship of the HSP22 and HSP27 clades, is shown in Fig. 1B. The mammalian HSP22 form a monophyletic group which is supported by the posterior probability of 1.0. The HSP22 clade is sister to the clade of HSP25/27 of Euteleostomi including mammals, birds and fish. Thus, HSP22 is most closely related to HSP27 among all proteins contained in the databases, while protein kinases or proteins related to the Herpex simplex protein kinase ICP10 were not identified.
Based on this analysis, HSP22 clearly is classified as a member of the superfamily of sHSPs. Therefore, we analyzed HSP22 for the property, which is characteristic for this superfamily: the ability to interact with itself and other sHSPs, which results in the formation of complexes.
Size Distribution of HSP22 Complexes in Monkey Heart-In order to determine the possible involvement of HSP22 in sHSP complex formation in tissues, a protein extract of a piece of Rhesus monkey heart was analyzed by gel filtration HPLC followed by SDS-PAGE and Western blotting. The proteins were extracted under moderate stringency conditions in order to leave sHSP complexes intact. HSP22 partitioned into all fractions corresponding to molecular masses between 25 kDa and Ͼ670 kDa with most of the proteins being found in the high molecular mass fractions (Fig. 2). For comparison, the size distribution of two further sHSPs, HSP27 and MKBP, was also analyzed. HSP27 showed a molecular mass partition, which is very similar to that of HSP22 and also to previous analyses (28), while MKBP partitioned into fractions with smaller molecular masses approximately between 25 and 200 kDa. These data suggest that (i) different sHSPs may have different size distributions, and (ii) HSP22 may interact with itself and/or other sHSPs, such as HSP27 and MKBP, which partition completely or partially into the same fractions.
HSP22-HSP22 Interaction-Earlier biochemical data suggested that HSP22 forms homo-dimers (1). Here we examined the HSP22-HSP22 interaction in greater detail. Yeast TH as- says using the constructs 1 and 2 (Table I) were performed to determine the interaction of HSP22 with itself (Fig. 3A). Both reporter genes (hisϩ, galϩ) clearly indicated interaction of HSP22 with itself. This interaction was verified biochemically by co-IP using Myc-and FLAG-tagged HSP22 (constructs 7, 8) expressed in COS-7 cells (Fig. 3B). As expected, after transfection with HSP22-myc the anti-FLAG antibody did not precipitate HSP22-myc, while after transfection with Flag-HSP22 the anti-FLAG antibody did precipitate Flag-HSP22 (controls). After co-transfection with both HSP22-myc and Flag-HSP22, the anti-FLAG antibody pulled down both Flag-HSP22 and HSP22-myc indicating that HSP22 interacts with itself. Under the same conditions, actin remained in the supernatants of the precipitation steps suggesting that the HSP22-myc precipitation is due to the interaction with Flag-HSP22 and not due to a nonspecific protein precipitation or trapping of proteins in the immunoprecipitate (Fig. 3B).
To determine the interacting domains in this HSP22 homo-dimer, two fragments of HSP22, each covering the N-(N) or the C-terminal (C) part of the molecule (Fig. 1A), were cloned into the TH vectors (constructs 3-6) and used for interaction assays which also included full-length (F) HSP22 (construct 1). HSP22-N interacted with HSP22-F, itself, and also with HSP22-C (Fig. 3C), while HSP22-C interacted with HSP22-F and HSP22-N, but not with itself (Fig. 3D). Thus, HSP22 appears to have two interacting domains potentially involved in homo-dimer formation (N-N, N-C interactions).
Next we examined whether HSP22 can form homo-oligomers, similar to what has been shown for HSP27 (22). 293T cells were transfected (construct 8) to express FLAG-tagged HSP22 and treated with different concentrations of glutaraldehyde to cross-link cell proteins. HSP22 species were analyzed by SDS-PAGE and Western blotting by using anti-FLAG antibody (Fig. 3E). CL resulted in three major HSP22 bands with apparent molecular masses of ϳ30, 65, and 120 kDa, while at higher glutaraldehyde concentrations a background signal (smear) with a molecular mass above 250 kDa was obtained. Distinct bands indicate CL of HSP22 with a specific binding partner rather than with a nonspecific variety of cell proteins, which would result in a continuous smear. Since 293T cells do not contain any sHSP, it is very likely that the major bands at 65 and 120 kDa represent cross-linked homo-dimers and homo-tetramers. The determined molecular masses are somewhat higher than the calculated molecular masses (22.6 kDa for the monomer) of HSP22, however, a similar retardation of electrophoretic mobility has also been observed for other sHSPs. Using higher concentrations of glutaraldehyde resulted in high molecular mass background signals (a broad smear) probably indicating nonspecific CL involving several cell proteins. For control purposes, 293T cells were transfected, but not treated with glutaraldehyde (Fig. 3E), or not transfected, but treated with glutaraldehyde (not shown). None of these treatments resulted in the formation of high-molecular mass HSP22 signals. As a further control, the same blot was developed for actin. Actin was not cross-linked by any of the used glutaraldehyde concentrations which indicates that CL of HSP22 is specific and not due to general CL of cell proteins (Fig. 3E). Taken together, these data provide evidence that HSP22 forms homo-dimers and homo-oligomers.
HSP22-cvHSP Interaction-TH assays using the constructs 2 and 25 were performed to determine the interaction between full-length HSP22 and cvHSP. After double transformation, both reporter genes (hisϩ, galϩ) were activated indicating interaction between the two proteins (Fig. 4A). This interaction was verified biochemically by co-IP (Fig. 4B). For co-IP, COS-7 cells were transfected with vectors coding for full-length Myctagged HSP22 (construct 7) and FLAG-tagged cvHSP (construct 28). The expressed proteins were analyzed in a manner similar to the experiment described above. After expression of both proteins, the anti-FLAG antibody pulled down both the Flag-cvHSP and HSP22-myc indicating that HSP22 interacts with cvHSP. Actin was not precipitated indicating that the HSP22-myc precipitation is due to the interaction with Flag-cvHSP and not due to nonspecific protein precipitation.
As a further method to study the interaction of both proteins in vivo we used FRET microscopy. COS-7 cells were co-transfected with vectors that permit expression of HSP22-YFP and cvHSP-CFP (constructs 9, 29) fusion proteins (Fig. 4C). Fluorescence intensity of the CFP signal (panels a and b) was determined before (panels a and c) and after (panels b and d) photobleaching of the YFP (panels c and d) signal in the region of interest. The fluorescence increase of the CFP signal was used to calculate the FRET Factor (FF) (0.24 Ϯ 0.024, p ϭ 0.0005), which is significantly different from the negative control (see "Experimental Procedures") indicating interaction between the two proteins in vivo.
HSP22-MKBP Interaction-TH assays using the full-length HSP22 and MKBP (constructs 2, 20) activated both reporter genes (hisϩ, galϩ) indicating interaction between the two proteins (Fig. 5A). This interaction was also verified biochemically by co-IP using full-length FLAG-tagged HSP22 and Myctagged MKBP (constructs 8, 23) (Fig. 5B). Similar to the experiments described above, after co-transfection with both Flag-HSP22 and MKBP-myc the anti-Myc antibody pulled down both proteins indicating that HSP22 interacts with MKBP. Again, actin was not precipitated suggesting that the Flag-HSP22 precipitation is due to a specific interaction with MKBP-myc.
For FRET measurements, COS-7 cells were co-transfected with vectors which permit expression of HSP22-YFP and MKBP-CFP (constructs 9, 24) fusion proteins (Fig. 5C). From the fluorescence intensity of the CFP signal (panels a and b) before (panels a and c) and after (panels b and d) photobleaching of the YFP (panels c and d) signal the FF (0.21 Ϯ 0.0024, p ϭ 0.0036) was calculated. This value is significantly different from the negative control (see "Experimental Procedures").

Thus both proteins interact in vivo.
We also sought to determine the interacting domains in the HSP22-MKBP hetero-dimer using the N-and the C-terminal halves of both proteins (constructs 5, 6, 21, 22; Fig. 1A), together with the full-length proteins (constructs 2, 20), in TH assays. HSP22-F did not interact with any fragment of MKBP (Fig. 5D). HSP22-N interacted with MKBP-F, but not with MKBP-N or MKBP-C (Fig. 5E). Finally, HSP22-C did not interact with MKBP-F, MKBP-N, or MKBP-C (Fig. 5F). These data suggest that HSP22-MKBP interaction involves the Nterminal half of HSP22, but appears to require full-length MKBP (N-F interaction).
HSP22-HSP27 Interaction-In a previous publication we reported that HSP22 (construct 2) interacts with a mimic of phosphorylated HSP27 ( 3D HSP27; construct 11), but not with wild-type HSP27 ( wt HSP27; construct 10) (1). In the present study we confirmed these data (Fig. 6A). However, when the vectors and sHSP inserts were exchanged in the TH assays (constructs 1, 15, 16), there was also an activation of both reporter genes when wt HSP27 cDNA was used indicating interaction between HSP22 and HSP27, regardless of the phosphorylation state of HSP27 (Fig. 6B). In order to resolve these conflicting TH data, a number of co-IP, CL and FRET experiments were performed. While the co-IP and CL experiments were unable to demonstrate interaction of HSP22 with any form of HSP27 (not shown), the FRET measurements did detect interaction. For FRET analysis, COS-7 cells were co-transfected with vectors that permit expression of HSP22-YFP (construct 9) and either wt HSP27-CFP, 3A HSP27-CFP or 3D HSP27-CFP (constructs 17-19) fusion proteins (Fig. 6, C-E, respectively). The 3A HSP27 construct permits the expression of a non-phosphorylatable HSP27 fusion protein in which all three MAPKAPK-2 phosphorylation sites in HSP27 were replaced by alanine residues. The fluorescence intensity of the CFP signal (panels a and b) was determined before (panels a and c) and after (panels b and d) photobleaching of the YFP (panels c and d) signals. The calculated FF for wt HSP27 (0.44 Ϯ 0.045, p Ͻ 0.0001), 3A HSP27 (0.32 Ϯ 0.047, p ϭ 0.0006), and 3D HSP27 (0.34 Ϯ 0.042, p Ͻ 0.0001) are all significantly different from the control (see "Experimental Procedures"). Thus, by FRET assay, HSP22 interacted with all forms of HSP27 regardless of the phosphorylation status of HSP27.

DISCUSSION
The biological role of HSP22 has been controversially discussed. While some groups considered HSP22 to be an sHSP, others classified it as a protein kinase homologous to the large subunit of Herpex simplex virus type 2 ribonucleotide reductase (ICP10) which also has a protein kinase activity (4 -6).
This analysis did not identify any eukaryotic, prokaryotic or viral protein kinase, nor any protein related to ICP10. Although sHSPs are known to have conserved a few protein kinase catalytic domains, which may be the basis for the observed autophosphorylation activities (4, 29), they are not known to be biologically significant protein kinases. At present there is no convincing biochemical evidence (e.g. K m -value, specific activity, stoichiometry of the enzymatic reaction, substrate requirements) available which would support a role of HSP22, ␣B-Cry, or any other sHSP as a protein kinase with biological significance.
With HSP22 being a member of the sHSP superfamily, we reasoned that in addition to interacting with HSP27 it may also interact with other sHSPs abundant in muscles. The HPLC analysis of the monkey heart protein extract shows that HSP22 forms high molecular mass complexes which may well result from such interactions. To determine the interaction of HSP22 with other sHSPs, we used several approaches including genetic (TH), immunological (co-IP), chemical (CL), and cell biological (in vivo FRET) methods. The use of several methods helps to minimize both false-positive and false-negative results, which are known pitfalls of each of these methods. We showed using different methods that HSP22 interacts with itself forming homo-dimers, which further dimerize to form tetramers. Thus, HSP22 shares the complex-forming characteristics, which are typical for the other studied sHSPs. Similarly, we showed by independent methods that HSP22 forms hetero-dimers with cvHSP and MKBP.
For HSP22 interaction with HSP27, the obtained TH data were conflicting. Originally, only 3D HSP27 (mimicking phosphorylation) but not wt HSP27 was found to interact with HSP22 using a certain TH environment (1). In the present study we confirmed this finding, which clearly suggests that the insertion of negative charges at the HSP27 phosphorylation sites causes conformational changes, which support interaction with HSP22. However, as shown in this study, when the sHSP inserts and TH vectors were exchanged, there was a comparable interaction of both 3D HSP27 and wt HSP27 with HSP22. Thus, the conformational change caused by substitution of serine by aspartate residues is not critical for interaction under these conditions. We conclude from these results that in such fusion proteins the Gal4(ϩ)-transcription activation and -DNA binding domains have an impact on the steric accessibility of the proteins to be tested in these binding assays. Indeed, similar asymmetric results in TH experiments have been observed by others (30).
We conducted a series of biochemical and cell biological experiments to further address the issue of whether HSP22 and HSP27 interact in a phosphorylation-dependent way. Using different approaches (including expression of wt HSP27, 3D HSP27, and HSP22 in COS-7 cells, followed by co-IP, native PAGE, native IEF, or CL), we did not observe evidence of interaction between these two sHSPs (not shown). However, we could demonstrate in vivo interaction between HSP22 and HSP27 by FRET using CFP/YFP-fusion proteins, regardless of whether the wt HSP27, 3A HSP27, or 3D HSP27 constructs were used. Thus, the available data suggest that phosphorylation of HSP27 does not contribute to its interaction with HSP22, though it causes a conformational change in HSP27, which can be detected in TH experiments under certain conditions. Based on multiple alignment, structural domains of sHSPs have been identified: the conserved ␣-crystallin domain, the less conserved N-terminal domain, the variable central region, and the variable region of the C-terminal tails (Ref. 24; Fig.  1A). A number of previous reports suggested the involvement of at least two binding sites, one in the ␣-crystallin domain and one in the N-terminal domain, in sHSP complex formation. In a TH study it was shown that the conserved C-terminal domain is essential for the interaction between ␣A-Cry, ␣B-Cry, and HSP27 subunits (31). For HSP27-HSP27 and HSP27-␣B-Cry interaction, the major interacting site in the rat HSP27 sequence was defined as being amino acid residues 141-176 (23). Also yeast HSP42p interacts with itself via a conserved Cterminal site (32). The isolated C-terminal domain of mammalian ␣A-Cry forms dimers or tetramers, while the isolated Nterminal domain was still able to form a high molecular mass complexes (33). Also for Caenorhabditis elegans HSP16 -2, the N-terminal domain was shown to be essential for oligomerization into high molecular mass complexes (34). Two binding sites were found to be involved in HSP27-HSP27 interaction: one in the ␣-crystallin domain (which was insensitive to phosphorylation), the other in the far N terminus (which was sensitive to phosphorylation of Ser 90 ) (22). It was proposed that HSP27 forms stable dimers through the ␣-crystallin domain, and these dimers then further multimerize through the phosphorylation-sensitive N-terminal domain. Based on these data we reasoned that HSP22 also may have two binding sites which are involved in interaction with other sHSPs. For that reason, we separated the cDNA of HSP22 and three of its binding partners, MKBP, cvHSP, and HSP27 (including phosphorylation site mutants) into N-and C-terminal parts and tested these fragments in TH assays. The data suggest that HSP22 has at least two binding sites, which have specificity in interaction with other sHSPs. HSP22-N interacts both with itself and with HSP22-C, while HSP22-C does not interact with itself. This N-N and N-C interaction would permit the formation of homo-dimers and homo-oligomers, as has been detected in transfected cells. In contrast to this interaction, HSP22 and cvHSP interact exclusively through their C termini (C-C interaction). The interaction between HSP22 and HSP27 involves the C terminus of HSP27, but apparently requires full-length HSP22. Similarly, the interaction of HSP22 and MKBP involves the N terminus of HSP22, but appears to require fulllength MKBP. At this time, it is not clear whether the lattermost interactions indeed require one full-length binding partner, or whether this results from false-negative TH data.
Taken together, the present study shows, by in vivo and in vitro approaches, that HSP22 has the ability to bind to itself and three other sHSPs (cvHSP, MKBP, HSP27) which are abundant in muscles. These results also indicate that HSP22, like the other studied sHSPs, has at least two binding domains, which bind specifically to defined sites of other sHSPs. In a forthcoming study, the interactions between HSP22 and the remaining muscle sHSPs (␣B-Cry, HSP20, HSPB3) will be described. 2 It is hoped that this detailed analysis of the sHSP interactions will contribute to a better understanding of the biological role of the sHSP complexes in muscles and other tissues.