Sequence Characteristics, Subcellular Localization, and Substrate Specificity of DYRK-related Kinases, a Novel Family of Dual Specificity Protein Kinases*

DYRK1 is a dual specificity protein kinase presumably involved in brain development. Here we show that the kinase belongs to a new family of protein kinases comprising at least seven mammalian isoforms (DYRK1A, DYRK1B, DYRK1C, DYRK2, DYRK3, DYRK4A, and DYRK4B), the yeast homolog Yak1p, and the Drosophila kinase minibrain (MNB). In rat tissues, DYRK1A is expressed ubiquitously, whereas transcripts for DYRK1B, DYRK2, DYRK3, and DYRK4 were detected predominantly in testes of adult but not prepuberal rats. By fluorescence microscopy and subcellular fractionation, a green fluorescent protein (GFP) fusion protein of DYRK1A was found to accumulate in the nucleus of transfected COS-7 and HEK293 cells, whereas GFP-DYRK2 was predominantly detected in the cytoplasm. DYRK1A exhibited a punctate pattern of GFP fluorescence inside the nucleus and was co-purified with the nuclear matrix. Analysis of GFP-DYRK1A deletion constructs showed that the nuclear localization of DYRK1A was mediated by its nuclear targeting signal (amino acids 105–139) but that its characteristic subnuclear distribution depended on additional N-terminal elements (amino acids 1–104). When expressed inEscherichia coli, DYRK1A, DYRK2, DYRK3, MNB, and Yak1p catalyzed their autophosphorylation on tyrosine residues. The kinases differed in their substrate specificity in that DYRK2 and DYRK3, but not DYRK1A and MNB, catalyzed phosphorylation of histone H2B. The heterogeneity of their subcellular localization and substrate specificity suggests that the kinases are involved in different cellular functions.

DYRK1 is a dual specificity protein kinase presumably involved in brain development. Here we show that the kinase belongs to a new family of protein kinases comprising at least seven mammalian isoforms (DYRK1A, DYRK1B, DYRK1C, DYRK2, DYRK3, DYRK4A, and DYRK4B), the yeast homolog Yak1p, and the Drosophila kinase minibrain (MNB). In rat tissues, DYRK1A is expressed ubiquitously, whereas transcripts for DYRK1B, DYRK2, DYRK3, and DYRK4 were detected predominantly in testes of adult but not prepuberal rats. By fluorescence microscopy and subcellular fractionation, a green fluorescent protein (GFP) fusion protein of DYRK1A was found to accumulate in the nucleus of transfected COS-7 and HEK293 cells, whereas GFP-DYRK2 was predominantly detected in the cytoplasm. DYRK1A exhibited a punctate pattern of GFP fluorescence inside the nucleus and was co-purified with the nuclear matrix. Analysis of GFP-DYRK1A deletion constructs showed that the nuclear localization of DYRK1A was mediated by its nuclear targeting signal (amino acids 105-139) but that its characteristic subnuclear distribution depended on additional N-terminal elements (amino acids 1-104). When expressed in Escherichia coli, DYRK1A, DYRK2, DYRK3, MNB, and Yak1p catalyzed their autophosphorylation on tyrosine residues. The kinases differed in their substrate specificity in that DYRK2 and DYRK3, but not DYRK1A and MNB, catalyzed phosphorylation of histone H2B. The heterogeneity of their subcellular localization and substrate specificity suggests that the kinases are involved in different cellular functions.
DYRK1 is a dual specificity protein kinase that catalyzes its autophosphorylation on serine/threonine and on tyrosine residues (1). The Drosophila homolog of this gene, minibrain, encodes three splicing variants of the protein kinase MNB. Mutant flies with a reduced expression of minibrain have a reduced number of neurons in distinct areas of the adult brain (2). In addition, mutant minibrain flies show specific behavioral defects (2,3).
Due to the high sequence similarity of the rat DYRK1 cDNA with a human expressed sequence tag (EST) 1 that had previously been mapped to chromosome 21, we were able to localize the gene of a human homolog of DYRK1 to 21q22.2 (1). By correlation of phenotype with genotype in patients with partial trisomies, this region has been defined as the "Down's syndrome critical region"; its triplication appears to be responsible for many features of Down's syndrome including mental retardation (4 -6). In sequencing projects of the Down's syndrome critical region, several groups have independently identified the human DYRK1 gene (7)(8)(9)(10)(11). Because of its high similarity with Drosophila MNB, DYRK1 is currently considered a candidate gene for the aberrant development of the brain that underlies mental retardation in Down's syndrome. The product of the human DYRK1 gene is nearly identical with rat DYRK1 (3 of 763 amino acids differ). Recently, Smith et al. (12) reported that transgenic mice with a 180-kb fragment of human chromosome 21 including the DYRK1 gene exhibit defects in learning tasks.
In Saccharomyces cerevisiae, Yak1p is the protein kinase with the highest sequence similarity with MNB and DYRK1. The YAK1 gene was identified as a functional antagonist of the RAS/protein kinase A pathway and has been characterized as a negative regulator of growth (13,14). These kinases share several sequence motifs in the catalytic domain that distinguish them from all other known protein kinases, e.g. a conserved sequence motif in the activation loop including putative regulatory tyrosine residues.
To define the family of DYRK-related kinases further, we report the identification and characterization of new members of the family. Our data indicate that DYRK-related kinases constitute a distinct family of protein kinases characterized by the structural similarity of their kinase domains and their capability to autophosphorylate on tyrosine residues. However, members of the DYRK family have unrelated sequences outside the catalytic domain and differ in their substrate specificity, tissue distribution, and subcellular localization.

RNA Preparation and PCR Cloning of DYRK-related Kinases-Total
RNA was prepared from differentiated 3T3-L1 cells (15) and from various rat tissues by the method of Chirgwin et al. (16). First-strand cDNA was synthesized with oligo(dT) as primer (First-strand cDNA Synthesis Kit, Amersham Pharmacia Biotech). Partial cDNAs of DYRK-related kinases were amplified by nested PCRs. Degenerate oligonucleotides corresponding with subdomains VIB and IX of the catalytic domain of serine/threonine kinases were used as primers in the first PCR (1). From the products of this reaction, partial cDNAs for DYRK-related kinases were amplified with the same forward primer and a reverse primer specific for a motif conserved in DYRK1, MNB, and Yak1p (amino acid sequence YIQSRFYR, oligonucleotide sequence 5Ј-TA(A/G)AA(A/C/G/T)C(G/T)(A/C/G/T)(C/G)(A/T)(C/T)TG(A/C/G/T)-AT(A/G)TA-3Ј). For amplification of DYRK4A cDNA, a gene-specific primer (5Ј-TCAGCACCATCGTGAGGTCC-3Ј) was used as the reverse PCR primer. PCR products were cloned into a plasmid vector and sequenced.
Library Screening and Sequencing-Cloned PCR products of two novel DYRK-related kinases were used as probes to screen a human fetal brain cDNA library (Stratagene). The sequences of the cDNA inserts were determined with overlapping fragments from both strands.
Northern Blot Analysis-Samples of total RNA (15 g) were separated by electrophoresis through denaturing 1% agarose gels containing 1% formaldehyde and transferred onto nylon or nitrocellulose membranes (Hybond N ϩ or Hybond C, Amersham Pharmacia Biotech). Equal loading of different samples was ascertained by ethidium bromide staining. Partial cDNA probes were labeled with [␣-32 P]dCTP by random oligonucleotide priming to specific activities of 0.5-1 ϫ 10 9 cpm/g (17). The following cDNA sequences were used as probes: DYRK1A, bp 125-1680 of the rat cDNA clone (1); DYRK2, bp 1344 -2154 of the human cDNA depicted in Fig. 2; DYRK3, complete human cDNA clone (Fig. 3). For DYRK1B and DYRK4A, the cloned murine PCR products were used (Fig. 1). Membranes were hybridized at 42°C in 50% formamide at a probe concentration of 0.5-1 ϫ 10 7 cpm/ml and subsequently washed three times at 55°C in 120 mM NaCl, 12 mM sodium citrate, 0.1% SDS. This degree of stringency was chosen to prevent cross-hybridization between cDNA probes and transcripts of DYRK1, DYRK2, DYRK3, and DYRK4 whose mRNA sequences are 58 -72% identical. Cross-reaction between DYRK1A/DYRK1B and DYRK4A/DYRK4B was not excluded.
Preparation of Recombinant Fusion Proteins-The expression plasmids for GST-DYRK1A and its mutated version GST-DYRK1A-K188R have been previously described (1). To create a GST-DYRK2 expression vector, a BamHI site was introduced at the presumed initiator codon of human DYRK2 cDNA by PCR. The modified DYRK2 cDNA (bp 386 -3499) was then ligated into pGEX-2TK (Amersham Pharmacia Biotech) via the engineered BamHI site and a HindIII site in the 3Ј-untranslated region. For expression of DYRK3, a BspHI/EcoRI fragment of the cDNA (bp 252-2070) was inserted into the SmaI site of pGEX-2TK. For expression of Yak1p, a fragment of the YAK1 gene encoding the catalytic domain (amino acids 338 -713 (13)) was amplified with specific primers from yeast genomic DNA. Restriction sites (BamHI and EcoRI) for cloning into pGEX-2TK were introduced by appropriate design of the PCR primers. An MBP-MNB fusion clone was prepared by DNA amplification with the MNB isoform B cDNA (2) as the template. With the 5Ј-end primer (5Ј-CACCGGGATATCATGGACC-3Ј), a single mutation (A/G) was introduced in front of the initial ATG codon to create an EcoRV site, and an EcoRV-EcoRI fragment of the amplified MNB cDNA was cloned into the pMAL-c plasmid (New England Biolabs) digested with StuI and EcoRI. The cloned PCR products of DYRK2, Yak1p, and MNB were verified by sequencing.

Expression of Green Fluorescent Protein (GFP) Fusion Proteins and Subcellular Fractionation of COS-7 Cells-cDNAs
with full open reading frames for rat DYRK1A and human DYRK2 were inserted into the expression vector pEGFP-C1 (CLONTECH) to generate the respective N-terminal fusion of the kinase with GFP. Deletion constructs of GFP-DYRK1A as shown in Fig. 7A were created by subcloning of cDNA fragments with the help of natural or PCR-generated restriction sites. All constructs were sequenced to ensure the accuracy of the reading frames and to verify the fidelity of the PCR.
For microscopic analysis, COS-7 and HEK293 cells were seeded on coverslips and transiently transfected either with the help of commercially available transfection reagents (lipofection with DAC-30 from Eurogentec, Belgium; non-liposomal transfection with FuGENE 6, Boehringer Mannheim) or by calcium phosphate precipitation. No effect of the transfection method on the subcellular localization of the GFP fusion proteins was observed. 24 h after transfection (48 h for GFP-DYRK1 1-104 ), cells were washed 2ϫ with phosphate-buffered saline (138 mM NaCl, 2.6 mM KCl, 10 mM Na 2 HPO 4 , 1.8 mM KH 2 PO 4 , pH 7.4), fixed 20 -30 min in 3% paraformaldehyde, washed again 2ϫ with phosphate-buffered saline and 1ϫ with H 2 O, embedded in fluoromount-G (Southern Biotechnology Associates), and visualized with a Zeiss Axiophot equipped for epifluorescence. 2-5 independent transfections were performed with each fusion construct, and at least two different samples were evaluated microscopically in each experiment.
For subcellular fractionation, COS-7 cells were transfected by calcium phosphate precipitation as described (18). 3 days after transfection, cells were washed with cold TMNS buffer (10 mM Tris-Cl, pH 6.8, 100 mM NaCl, 3 mM MgCl 2 , 300 mM sucrose) and scraped off the culture dishes. Combined cells from two 90-mm plates were resuspended in cold TMNS supplemented with proteinase and phosphatase inhibitors (1 mM phenylmethylsulfonyl fluoride, 2 g/ml leupeptin, 1 g/ml pepstatin, 1 mM vanadate) and disrupted in a Teflon glass potter (720 rpm, 1 min), and nuclei were sedimented by centrifugation (600 ϫ g, 4°C, 5 min). The resulting supernatant was centrifuged at 100,000 ϫ g and 4°C for 75 min, resulting in a supernatant of soluble proteins and a pellet containing membranes and any insoluble material that escaped sedimentation in the first centrifugation step, e.g. partially disrupted nuclei. The 600 ϫ g pellet was resuspended in TMNS buffer supplemented with 1% (w/v) Tween 20 and layered over a sucrose cushion (1.12 M sucrose, 20 mM Tris-Cl, pH 7.4, 1 mM EDTA); nuclei were then sedimented at 45,000 ϫ g (4°C, 60 min). Alternatively, nuclei were fractionated into the insoluble nuclear matrix and a soluble chromatincontaining fraction (19,20). The 600 ϫ g pellet was resuspended in 2 ml of digestion buffer (10 mM Tris-Cl, pH 6.8, 50 mM NaCl, 3 mM MgCl 2 , 300 mM sucrose, 0.5% (v/v) Triton X-100) supplemented with proteinase inhibitors as above and RNase A and DNase I (each at 100 g/ml). After 30 min of incubation at room temperature, 1 M (NH 4 ) 2 SO 4 was added dropwise to a final concentration of 230 mM. Nuclear matrices were pelleted by centrifugation at 600 ϫ g, 4°C, for 5 min. As estimated by comparison of Coomassie-stained histone bands, 70 -90% of the chromatin was extracted from the nuclei by this method. Aliquots of the protein samples corresponding to an equivalent portion of the original cell homogenate were separated by SDS-PAGE, transferred onto PVDF membranes, and probed with a GFP-specific antiserum (CLONTECH). Reacting bands were detected using a peroxidase-based chemiluminescence kit (Boehringer Mannheim).
Assay of Tyrosine Autophosphorylation-Tyrosine autophosphorylation of recombinant fusion proteins was detected by Western blot analysis with an anti-phosphotyrosine antibody (PY20, Transduction Laboratories). To normalize for equal loading, the GST fusion proteins were quantified by Western blotting and subsequent immunochemical detection with a GST-specific antibody and 125 I-labeled protein A. The amount of MBP-MNB loaded onto the gel was adjusted by visual comparison of Coomassie-stained bands.
In Vitro Assay of Protein Kinase Activity-Samples of the recombinant kinases were incubated in phosphorylation buffer (25 mM Hepes, 5 mM MgCl 2 , 5 mM MnCl 2 , 0.5 mM dithiothreitol) containing 10 M [␥-32 P]ATP (100 Ci/ml) for 11 min (GST-DYRK1A and GST-DYRK2) or 30 min (GST-DYRK3 and MBP-MNB) at room temperature. Substrates were added to a final concentration of 8 M (histones H3 and H2B, Boehringer Mannheim) or 0.25 mg/ml (histone type II-S, catalog number H6005, Sigma; 0.1 mg/ml in assays with DYRK1A and DYRK2). Reaction products were separated by SDS-PAGE (16% gels) and visualized by autoradiography of the dried gels. The time of exposure of the x-ray films was adjusted to give comparable band intensities with the different kinases.

Identification of DYRK-related Kinases by PCR Cloning-
Based on a comparison of DYRK1, MNB, and Yak1p, a PCR cloning approach was designed to identify related kinases. Partial cDNAs of DYRK-related kinases were amplified from murine 3T3-L1 cells in two successive PCRs with nested primers. In the first PCR, degenerate oligonucleotide primers matching subdomains VIB and IX of protein serine/threonine kinases were used. In the second PCR, the same forward primer matching subdomain VIB was combined with a reverse primer corresponding with the YIQSRFYR motif before subdomain VIII that is fully conserved in DYRK1, MNB, and Yak1p (1). In addition, a reverse primer corresponding with the sequence of a human EST was used (GenBank TM accession number T93981). PCR products were cloned and characterized by sequencing.
As illustrated in Fig. 1, cDNA sequences encoding seven different DYRK-related kinases were cloned. Six of the cDNAs encoded protein kinases with a sequence motif specific for DYRK-related kinases (DYRK, MNB, and Yak1p), i.e. the amino acids "Ser-Ser-Cys" following subdomain VII. The designation of the kinases was based on the different degree of similarity among the sequences. Four kinases that shared less than 70% identical amino acids in this region were designated DYRK1, DYRK2, DYRK3, and DYRK4. Two pairs of closely related sequences (more than 95% of identical amino acids) were classified as isoenzymes (DYRK1A/B and DYRK4A/B).
DYRK1A corresponds with the rat and murine "DYRK" as described previously (1,9). The amino acid sequence of DYRK1B is identical with DYRK1A in this region, whereas the cDNA sequences are only 78% identical. In addition to DYRK1A and DYRK1B, the partial sequence of a third murine DYRK1 isoform was identified in a data base search (DYRK1C, GenBank TM accession number U49952). This cDNA sequence is 80% identical to DYRK1A and 88% identical to DYRK1B. Except for this data base entry, we found no sequence encoding a DYRK-related kinase that was not detected in our screen.
One of the cloned kinases differed from DYRK1-4 in the length of the spacer between subdomain VIB and VII and did not contain the characteristic Ser-Ser-Cys motif. Except for one amino acid exchange, its sequence is identical with that of a human protein kinase, PKY, expressed in multidrug-resistant cell lines (21). By sequence similarity of their catalytic domains, DYRK-related kinases are the closest relatives of PKY (37-40% of identity). However, PKY lacks some characteristic features of DYRK, MNB and Yak1p, e.g. a cysteine in subdomain VIB and the Tyr-X-Tyr motif in the activation loop.
Structural Characteristics of DYRK2 and DYRK3-Cloned PCR products of DYRK2 and DYRK3 were used to isolate full-length cDNA clones from a human fetal brain cDNA library. Seven cDNA clones of DYRK2 differing in the length of the 3Ј-untranslated region were isolated. In three clones, the poly(A) tail began only 184 bases after the stop codon (Fig. 2), whereas 4 of the cDNAs contained 1461 additional nucleotides of 3Ј-untranslated region. The existence of additional transcripts with an even longer 3Ј-untranslated region can be predicted from the sequences of four ESTs that were identified in the data base (GenBank TM accession numbers R63190, R63622, R25341, and Z25301). Furthermore, of three cDNA clones that contained the 5Ј-end of the transcript, one differed from the others by the presence of a 149-bp insertion (Fig. 2). Due to the shift of the reading frame, the protein encoded by this cDNA clone may start at a different initiator codon and thus contain 73 additional amino acids at its N terminus.
The deduced amino acid sequence of DYRK2 appears to be identical with a human kinase that has previously been presented as a partial sequence (PSK-H2) in an alignment of catalytic domains of kinases (22). Differences of 5 amino acids may represent errors in the sequence of PSK-H2. However, in the absence of information on the nucleotide sequence of PSK-H2, it cannot be excluded that PSK-H2 is encoded by a different gene than DYRK2.
A 2170-bp cDNA clone of DYRK3 was isolated that comprised an open reading frame of 550 codons and a poly(A)-tail (Fig. 3). There is no in-frame stop codon before the first ATG codon. However, the sequence preceding this putative initiator codon appears "non-coding" as judged from the unusually high proportion of rare codons when compared with "average" human proteins. According to the codon usage table for human proteins (23), 9 codons (11%) in the first 252 nucleotides were considered "rare," in contrast to only 20 in the 550 codons of the open reading frame starting with nucleotide 253 (3.6%; rare codon threshold 0.1). We therefore conclude that one of the two consecutive ATG codons at nucleotides 253-258 (Fig. 3) is most likely the initiator codon of DYRK3. Recently, a partial cDNA sequence of human DYRK3 was published as a data base entry (GenBank TM accession number U69558). This clone contains an insertion of 95 nucleotides relative to the clone presented in Fig. 3 and most probably represents an alternatively spliced version of the transcript. This longer transcript contains an ATG codon that initiates an open reading frame encoding 15 additional N-terminal amino acids relative to the sequence shown in Fig. 3.
Both DYRK2 and DYRK3 contain a canonical kinase domain (22) that is located between a large N-terminal domain (149 and 173 amino acids, respectively) and a short C-terminal extension (66 amino acids). The sequences exhibit the specific features of DYRK-related kinases (DYRK1, MNB, and Yak1p), including the motif Ser-Ser-Cys (SSC) in subdomain VII, a cysteine residue in subdomain VIB, and the conserved sequence in the activation loop (YXYIQSRFY, Fig. 2 and Fig. 3).
DYRK2 and DYRK3 are related over their entire sequences except for the first 100 amino acids of the N-terminal region (72% of identical amino acids). In contrast, their similarity with DYRK1 is restricted to the catalytic domain (46 and 43% of identity, respectively) and a small region of 20 amino acids immediately preceding the catalytic domain. DYRK2 and DYRK3 lack the striking sequence motifs that were identified in DYRK1, e.g. a nuclear targeting sequence, a PEST region, and a stretch of 13 consecutive histidine residues (1).
Genomic Localization of the Human Genes for DYRK2 and DYRK3-Data base searches identified a number of ESTs for DYRK2 and DYRK3. One of the ESTs for DYRK2 (GenBank TM accession number Z25301) has been mapped to human chromosome 12 in the Genexpress program (24). For DYRK3, the map position of an EST clone of the I.M.A.G.E. consortium (clone 23329; GenBank TM accession number R38268 (25)) was determined by fluorescence-based in situ hybridization to be 1q32 (26). The gene for the Van der Woude syndrome, an autosomal dominant craniofacial disorder, has been mapped to this region of chromosome 1 by linkage analysis (27).

Expression of DYRK-related Kinases in Different Rat
Tissues-Tissue distribution of the transcripts for DYRK-related kinases was studied by Northern blot analysis. As previously observed (1), a specific probe for DYRK1A hybridized with two bands of approximately 2.8 and 5.4 kb which represent alternatively polyadenylated mRNA species (Fig. 4A). DYRK1A is ubiquitously expressed in all rat tissues studied, although mRNA levels varied widely (weak bands in RNA from liver were detected in parallel experiments). In contrast, the probes specific for DYRK1B, DYRK2, and DYRK4 hybridized only with RNA from testis. After hybridization with the DYRK3 cDNA, very weak signals were found in organs other than testis (spleen and adrenal gland). All detected bands corresponded reasonably well with the transcript sizes of the known cDNA clones. ). An insertion found in one cDNA clone is shown in lowercase letters, and the resulting N-terminal extension of the protein sequence is given in thin print. Additional sequence information of the 3Ј-untranslated region in the alternatively polyadenylated transcripts has been submitted to the EMBL data base (accession number Y13493).
We also studied expression of DYRK-related kinases in testes of prepuberal (2 weeks of age), puberal (8 weeks), and mature rats (16 weeks). As shown in Fig. 4B, low levels of the large transcript of DYRK1A (arrow) were detected that did not change during puberty. A strongly hybridizing transcript of about 3 kb (asterisk) appeared only in puberal and adult rats; this band most likely represents the cross-hybridizing mRNA of DYRK1B. The transcripts for DYRK1B, DYRK2, DYRK3, and DYRK4 were detected only after onset of spermatogenesis.
Subcellular Localization of DYRK1A and DYRK2-DYRK1A harbors a bipartite nuclear localization signal in its N-terminal region (1), whereas no targeting signal was identified in DYRK2. In order to evaluate the functional significance of this difference, we expressed GFP fusion proteins of DYRK1A and DYRK2 in two different mammalian cell lines (COS-7 cells and HEK293 cells). The subcellular localization of the recombinant proteins was determined by fluorescence microscopy of intact cells (Fig. 5) and by biochemical fractionation of transfected COS-7 cells (Fig. 6). In both cell lines, GFP-DYRK1A was found to be present in the nucleus (Fig. 5). The nuclei were not stained homogeneously in that the fluorescent signal was excluded from the nucleolus and appeared concentrated in discrete, irregularly shaped speckles. Consistent with the microscopic evaluation, GFP-DYRK1A was found mainly in the nuclear fraction (600 ϫ g pellet) of transfected COS-7 cells (Fig.  6). In contrast, GFP-DYRK2 transiently expressed in COS-7 cells was not restricted to the nucleus. Moreover, in a small fraction of COS-7 cells (Ͻ2%, not shown in Fig. 5) and in the majority of HEK293 cells transfected with GFP-DYRK2, staining was largely excluded from the nucleus. Furthermore, after biochemical fractionation of COS-7 cells GFP-DYRK2 could not be detected in the nuclear fraction (Fig. 6). Possibly, the ab- sence of GFP-DYRK2 from the nucleus of COS-7 cells could not be visualized by conventional (versus confocal) microscopy because of the strong cytoplasmic fluorescence produced by the high levels of protein expression in this cell line. As observed by others (28,29), control transfections with the empty vector resulted in a strong staining of the entire cells.
Subnuclear Distribution of DYRK1A-A series of deletion constructs of GFP-DYRK1A was prepared in order to identify the domain(s) of DYRK1A that direct nuclear localization and confer its distinct subnuclear distribution (Fig. 7A). As shown in Fig. 7B, a GFP fusion protein containing the N-terminal region of DYRK1A (GFP-DYRK1A 1-176 ) exhibited exactly the same punctate pattern of nuclear staining as GFP-DYRK1A. A short fragment of DYRK1A containing the bipartite nuclear localization signal (DYRK1A 105-139 ) was sufficient to direct the fusion protein to the nucleus. In contrast to GFP-DYRK1A, this construct appeared homogeneously distributed in the nucleus with no indication of a specific subnuclear localization. A construct comprising amino acids 1-104 of DYRK1A was only slightly enriched in the nucleus but appeared to be excluded from nucleolus similar to GFP-DYRK1A (see also next paragraph). This result suggests that GFP-DYRK1A 1-104 is able to bind to specific intranuclear target structures once it enters the nucleus by random diffusion due to its small size (30).
The subcellular and subnuclear localization of the GFP-DYRK1A deletion constructs was also studied by biochemical fractionation of transfected COS-7 cells. To characterize the subnuclear distribution of DYRK1A, a "nuclear matrix" fraction was prepared by treatment of nuclei with a nonionic detergent, digestion with nucleases, and extraction with a buffer of high ionic strength (19 -21) (Fig. 8A). By this method, most of the chromatin and all soluble and loosely bounds proteins ("chromatin" fraction) are removed from an insoluble fraction ("matrix") containing the peripheral nuclear lamina and an intranuclear skeletal network. As shown in Fig. 8B, GFP-DYRK1A and GFP-DYRK1A 1-176 were found in the nuclear matrix, with no signal being detected in the soluble fractions (S, Ch). Reacting bands in the insoluble cytoplasmic fraction (100,000 ϫ g pellet, P) probably represent contaminating nuclear proteins, since the first centrifugation step at 600 ϫ g favors purity of the nuclear fraction rather than complete sedimentation of nuclei. In contrast, the majority of GFP-DYRK1A 105-139 was found in the soluble fractions. This result indicates that GFP-DYRK1A 105-139 lacks the domain necessary for the interaction of DYRK1A with the nuclear matrix, corresponding with the lack of the punctate subnuclear distribution of GFP fluorescence in COS-7 cells transfected with this construct (Fig. 7B). GFP-DYRK1A 105-139 was mainly found in the "cytoplasmic" fractions (600 ϫ g supernatant, S and P in Fig.  8), although its nuclear localization was clearly demonstrated by fluorescence microscopy (Fig. 7B). As observed with other proteins, this lack of correlation between data obtained by microscopy and cell fractionation is likely to result from leakage of soluble proteins during fractionation (32,33). Interestingly, GFP-DYRK1A 1-104 essentially resembles GFP-DYRK1A in its pattern of distribution. This result suggests that this region of DYRK1A (amino acids 1-104) mediates its association with an insoluble target structure, i.e. the nuclear matrix. Taken together, there is a strong correlation between the copurification of a GFP-DYRK1A construct with the nuclear matrix fraction (Fig. 8) and its association with the irregularly shaped subnuclear speckles visualized by fluorescence microscopy (Fig. 7B).
Protein Kinase Activity of DYRK-related Kinases-To characterize the enzymatic properties of DYRK-related kinases, we  (Fig. 1A), transcripts hybridizing with the human DYRK4A probe may represent either one or both isoforms. Migration of 18 S rRNA (1.9 kb) and 28 S rRNA (4.9 kb) is marked at the right. B, two different bands hybridizing with the DYRK1A probe are marked by an arrow and an asterisk (see text). EtBr, ethidium bromide staining of a gel loaded with the same RNA samples as the blots. 86 kDa) and a fragment of 46 kDa. MBP-MNB was detected as a single band of 103 kDa which also appears to represent a truncated product (calculated molecular mass, 132 kDa). The GST-Yak1p fusion protein migrated at the expected position (67 kDa). Fig. 9 illustrates that all recombinant proteins contained phosphotyrosine as detected with a specific antibody. A catalytically inactive point mutant of DYRK1A (DYRK-K188R (1)) did not react with the antibody, indicating that tyrosine phosphorylation of the recombinant kinases reflects indeed their intrinsic tyrosine kinase activity. Fig. 10 illustrates the protein kinase activity of the recombinant proteins toward different preparations of histone in vitro. GST-DYRK1A, GST-DYRK2, GST-DYRK3, and MBP-MNB catalyzed the incorporation of 32 P into histone H3 and histone type II-S, a fraction of calf thymus histones commercially available from Sigma. In contrast, histone H2B was only phosphorylated by DYRK2 and DYRK3 but not by DYRK1A and MNB. As determined by thin layer chromatography of phosphoamino acids, all histones phosphorylated in these experiments contained phosphoserine/threonine with no phosphotyrosine being detectable (data not shown). This result is discordant with our previous report that DYRK1A catalyzed phosphorylation of tyrosine residues in histone IIS (1). In these experiments (Figs. 5D and Fig. 6 in Ref. 1), a radioactively labeled product of hydrolysis was misidentified as phosphotyrosine. Under improved conditions of phosphoamino acid analysis, this unidentified molecule was clearly separated from the phosphotyrosine standard as well as from phosphotyrosine in hydrolysates of histone phosphorylated by the insulin receptor tyrosine kinase (data not shown). Thus, at present there is no evidence that DYRK1A has tyrosine kinase activity toward exogenous substrates.
We failed to detect in vitro kinase activity of GST-Yak1p toward histone, casein, or in autophosphorylation reactions (data not shown), although this construct was obviously able to catalyze tyrosine autophosphorylation in E. coli (Fig. 9). Thus, GST-Yak1p might have lost its enzymatic activity during the partial purification of the protein. It should be noted that an equivalent DYRK1A construct, containing only the catalytic domain fused to GST, is still an active protein kinase. 2

DISCUSSION
The human DYRK1A gene on chromosome 21 has received considerable attention because of its potential role in the mental retardation of Down's syndrome (7)(8)(9)(10)(11)(12)34). It is shown here that DYRK1A is a member of a small family of protein kinases that comprises at least seven different genes in mammals. These kinases exhibit characteristic common structural and biochemical features as well as striking functional differences. In particular, DYRK1A differs from other members of the family by three criteria as follows: 1) ubiquitous expression versus predominant expression in testis (DYRK1B and DYRK2-4); 2) Nuclei (N) were pelleted from cell homogenates at 600 ϫ g und further purified by sedimentation through a 1.12 M sucrose cushion. The postnuclear supernatant was separated into a soluble fraction (100,000 ϫ g supernatant, S) and an insoluble fraction (100,000 ϫ g pellet, P). Protein samples corresponding to the same fraction of the cell homogenate were separated by SDS-PAGE (16% gel and 8% gel as indicated) and blotted onto PVDF membranes. Recombinant proteins were detected immunochemically with a GFP-specific antiserum and peroxidase-mediated chemiluminescence. localization in specific subnuclear structures versus cytoplasmic localization (DYRK2); and 3) substrate specificity in histone kinase assays (DYRK2 and DYRK3).
The most striking common biochemical property of the DYRK kinases is their capability to autophosphorylate tyrosine residues (Fig. 9). Conservation of this feature in kinases from mammals, Drosophila, and yeast suggests that tyrosine autophosphorylation is important for the function of DYRK-related kinases. Toward exogenous substrates, e.g. histone (Fig. 10), the DYRK-related kinases exhibited different substrate specificity. Although the physiological substrates are not yet known, it is likely that DYRK1A/MNB and DYRK2/DYRK3 phosphorylate different substrates in vivo and may thus control different cellular processes.
Strong evidence for different cellular functions of DYRK1A and DYRK2 is also provided by their different patterns of subcellular distribution. A GFP-DYRK1A fusion protein was found in the nucleus of transfected COS-7 or HEK293 cells, whereas GFP-DYRK2 was localized in the cytoplasm. It has been shown by Song et al. (35) that a GFP fusion protein similar to GFP-DYRK1A 1-176 (Fig. 7A) is targeted to the nucleus of transiently transfected NIH 3T3 cells. This result was confirmed and extended by our analysis of GFP-DYRK1A deletion constructs. The presumed nuclear localization signal is indeed sufficient to direct a GFP fusion protein to the nucleus. Furthermore, the present data show that GFP-DYRK1A is associated with a distinct subnuclear compartment, microscopically visualized as irregularly formed speckles. In cell fractionation experiments, GFP-DYRK1A was co-purified with the nuclear matrix, originally defined by Berezney and Coffey (19,20) as the insoluble skeletal framework within the nucleus. This specific subnuclear localization of DYRK1A was not conferred by the nuclear localization signal alone but depended on the presence of the N-terminal region of the protein (amino acids 1-104). Similar punctate patterns of subnuclear distribution have previously been shown for other proteins, and several types of subnuclear compartments have been defined by their content of specific proteins, including the BRCA1 nuclear dots, PML bodies, and nuclear speckles containing the splicing factor SC-35 (36 -38). Among these, the SC-35-containing nuclear speckles and PML bodies appear to be clearly distinguishable from the subnuclear domains stained by GFP-DYRK1A simply by their smaller number (20 -50 speckles and 10 -20 PML bodies per nucleus (31) versus Ͼ100 GFP-DYRK1A-stained dots per nucleus). Thus far, the association of DYRK1A with a distinct subnuclear structure does not provide direct evidence for its involvement in any particular nuclear process. However, our data clearly point to a specific function of DYRK1A in the nucleus. MNB and DYRK1B 3 are likely to serve similar nuclear functions as DYRK1A since these kinases share considerable sequence similarity over 75 amino acids of their N-terminal regions, including the nuclear localization signal (83% of iden-tity with amino acids 83-157 of DYRK1A).
In addition to substrate specificity and subcellular distribution, DYRK-related kinases differ strikingly in their tissue distribution. Preponderant expression of MNB has been found in Drosophila embryonic and larval brain. Several authors have reported that mRNA of DYRK1A is present in all human, murine, and rat tissues studied (1,(7)(8)(9). In contrast, all other kinases identified in the present study (DYRK1B are DYRK2-4) are predominantly expressed in the testes of adult rats. Transcripts for the new DYRK-related kinases were not detectable in testes of prepuberal rats, suggesting that these kinases are involved in spermatogenesis. It should be noted that their transcripts are not strictly testis-specific, since we have isolated cDNA clones for DYRK2, DYRK3, and DYRK4 (but not for DYRK1B) 4 from a human fetal brain cDNA library. Furthermore, the dbEST data base (39) contains a considerable number of expressed sequence tags for DYRK2-4 that were isolated from other predominantly fetal tissues (brain, liver/ spleen). Thus, in addition to their function in testis, kinases of the DYRK family may play a role in embryonal development.
Sequence comparisons of DYRK1, DYRK2, DYRK3, MNB, and Yak1p allowed a definition of the common structural features of the DYRK family. Throughout the catalytic domain, members of the DYRK family share characteristic residues that are rarely found in other kinases (highlighted by shaded boxes in Figs. 2 and 3). The most striking common motifs are located in the core of the catalytic domain as follows: the conserved sequences of subdomains VI (HCDLKPEN), VII (DFGSSC), and VIII (YXYIQSRFYR(S/A)PE, where X is Q, H, or T). Mutational analysis has shown that tyrosines in the latter motif (Y 319 XY 321 in DYRK1A) are essential for the activity of DYRK1A (1). In analogy with the mitogen-activated protein kinases, these tyrosines are likely to represent the site of an activating phosphorylation of DYRK1A. Since all DYRKrelated kinases share the YXY motif in the activation loop, we suggest that their activity is regulated through a similar mechanism. Outside the catalytic domain, only a small region (20 amino acids) preceding the catalytic domain is conserved in the DYRK family. These sequence motifs are found in all members of the DYRK family that are presently accessible in the data base, including several sequences that were recently identified in genome sequencing projects of fission yeast (Schizosaccharomyces pombe), the roundworm Caenorhabditis elegans, and Drosophila melanogaster (GenBank TM accession numbers Z54354; Z50142 and Z69885, Z70308, L43478, and L43480, respectively). In addition, two ESTs from Arabidopsis thaliana (accession numbers N65563 and R84142) apparently encode protein kinases with the Ser-Ser-Cys motif and a cysteine in 3 S. Leder and W. Becker, unpublished data. 4 S. Leder and W. Becker, unpublished data.  Five different kinases of the DYRK family were expressed in E. coli as fusion proteins with either GST (DYRK1A, DYRK2, DYRK3, Yak1p) or MBP (MNB). To control for antibody specificity, a catalytically inactive mutant of DYRK1A (GST-DYRK1A-K188R) was analyzed in parallel. Fusion proteins were partially purified by affinity adsorption and separated by gel electrophoresis (8% acrylamide gel). Phosphotyrosine was detected immunochemically with a specific monoclonal antibody and 125 I-protein A. subdomain VIB. The conservation of DYRK-related kinases in plants, fungi, and animals suggests that these enzymes regulate fundamental functions of the eukaryotic cell.
In view of the interest focused on the DYRK1 gene on human chromosome 21, it is noteworthy that there are three very similar murine isoforms of DYRK1 (DYRK1A, -B, and -C). cDNA fragments for DYRK1B were also amplified from human testis, 2 whereas the existence of a third isoform of DYRK1C can presently only be inferred from the data base entry of a partial murine cDNA clone (Fig. 1). The presence of multiple isoforms of DYRK1 points to an important role of this enzyme. However, the different patterns of expression of DYRK1A and DYRK1B indicate that these kinases are not functionally redundant but may play similar roles in the regulation of nuclear functions in different organs.