INTRODUCTION

Determination of the positions of proteins within the 30 S and 50 S ribosomal subunits has been a major goal of studies directed toward the elucidation of the structure and function of the Escherichia coli ribosome. The technique of immune electron microscopy (IEM),¹ the visualization in electron micrographs of complexes formed between ribosomal subunits and specific antibodies, has been especially useful for this purpose (1, 2). However, in the case of protein L23, application of IEM has yielded controversial results. The object of the present work is to resolve this controversy.

E. coli ribosomal protein L23 is a primary binding protein; it interacts directly with ribosomal RNA (3) and plays a significant role in the assembly of the large ribosomal subunit (4). The protein has been linked to the ribosomal peptidyltransferase center in several ways. First, photoaffinity labeling of 70 S ribosomes with either puromycin (5-7) or p-azidopuromycin (8, 9), each of which inhibits protein synthesis by acting as a peptide acceptor in the transferase reaction, led to photoincorporation into L23 as the major site of protein labeling. Second, antibodies recognizing the N⁶,N⁶-dimethyladenosine moiety of puromycin bound to labeled 50 S subunits in a region generally agreed to include the peptidyltransferase center, i.e. between the central protuberance and the site of protein L1, near the 30 S:50 S interface (10, 11). Third, chemical cross-linking and related studies showed L23 to neighbor other proteins (L2, L15, L16, and L27) that have been placed at or near the peptidyltransferase center (12). Finally, although peptidyltransferase activity is not altered in reconstituted 50 S subunits either lacking L23 or including puromycin-modified L23 in place of L23, the latter do show reduced aminoacyl-tRNA binding (6), which suggests proximity to the transferase region.

In contrast, the Berlin group used IEM with polyclonal antibodies to place the protein far from the transferase center, at the subunit base on its cytoplasmic surface (13) at a position similar to that at which the growing peptide emerges from the subunit (the exit site) (14). IEM of subunits that had been photoaffinity modified with puromycin and azidopuromycin also shows a small but consistent secondary puromycin localization, at the exit site (10, 11). Furthermore, some cross-linking studies placed L23 near proteins (L29, L34) located near the 50 S base (15, 16). The peptidyltransferase and exit sites are separated by more than 100 Å in common subunit models (13, 17, 18).

As described previously (12), these conflicting results could be explained by one of three possibilities. First, the IEM placement of puromycin may not reflect placement of L23. We considered this unlikely because 70% of photoincorporation of puromycin into 50 S subunits was into L23. Second, the IEM placement using anti-L23 may be in error, either because the anti-L23 preparation is contaminated with antibodies to other ribosomal proteins or because an incorrect site has been identified. However, the authors argue persuasively for the validity of their results. Third, the IEM results cited above may all be correct; the puromycin binding site on L23 and epitopes recognized by anti-L23 IgG could differ and involve separate areas of an asymmetric protein. It is also possible that the two sites are not as far apart as is suggested by the visually derived subunit models. Reconstructed images of negatively stained subunits (19, 20) and of frozen unstained ribosomes (21-23) and electron microscopy of crystalline ribosome arrays (24) each show a more porous particle than the "classical" models of Lake (18), Vasiliev (17), or Stöffler (13) would suggest.

Here we report the definitive localization of L23 within the 50 S subunit by reconstitution of 50 S subunits with dinitrophenyl (DNP) derivatives of L23 in place of L23 and visualization of the complexes that such subunits make with anti-DNP. This approach, applied by us earlier to the localization of proteins within the 30 S subunit (25-28), avoids the possible ambiguities associated with the earlier L23 studies. In particular, derivatizing highly purified L23 with dinitrofluorobenzene ensures that anti-DNP will recognize only L23 in the reconstituted 50 S subunit. Our results clearly support the validity of both of the earlier IEM studies (i.e. of the third possibility) with interesting implications for 50 S structure.

EXPERIMENTAL PROCEDURES

Buffers and reagents were prepared and obtained essentially as described previously, as were E. coli Q13 70 S ribosomes and 50 S subunits (6, 26, 28). RP-HPLC was performed on both analytical (250 mm × 4.6 mm) and preparative (250 mm × 10 mm) Synchropak RP C18 and C8 columns (SynChrom Inc.) as described previously (6, 26, 28) and in the figure legends.

TP50 was prepared from 50 S subunits as described (6). RP-HPLC fractionation of TP50 on a preparative C18 column afforded two pools: L22/L23/L29 (fractions eluting between 52 min and 62 min in Fig. 1A) and TP50-L22/L23/L29 (all other protein-containing fractions). The L22/23/29 pool was further fractionated on an analytical C18 column (Fig. 1B). The peak L22 fraction, eluting at 42 min, afforded reasonably pure L22 (Fig. 2B). The L23/L29 pool derived from this chromatography (Fig. 1B, fractions eluting 45-50 min), still containing small amounts of L22, was further resolved on an analytical C8 column (Fig. 1C). The L29 sample derived from this column, slightly contaminated with L22 and L23, was used in reconstitution. The L23 sample was further purified by chromatography on an analytical C18 column (Fig. 1D), from which only the center fractions were retained. This last step was repeated twice to remove any trace of L22 and L29 (Fig. 2). Each protein was independently identified by a combination of PAGE analyses and A₂₈₀/A₂₁₅ ratio (29). Purified proteins L22 and L29 were combined with TP50-L22/L23/L29 to give TP50-L23. Protein amounts in protein pools (TP50, TP50-L23, and TP50-L22/L23/L29) were estimated by A₂₃₀ (30). Amounts of single proteins (L22, L23, or L29) were estimated by A₂₁₅ (31).

RNA for reconstitution was prepared by three phenol extractions of 50 S subunits, followed by a chloroform extraction of the aqueous layer, essentially as described (6). rRNA samples were analyzed by 3-5% SDS-PAGE to monitor both 23 S rRNA degradation and the presence of 5 S rRNA. The rRNA was dissolved in TM4 buffer (10 mM Tris·HCl, pH 7.4, 4 mM magnesium acetate, 50 mM KCl) for use in reconstitution experiments.

Protein L23 was modified by incubation with 0.14 M [³H]DNFB (DuPont NEN; diluted with nonradioactive DNFB to a specific radioactivity of 10 Ci/mol) by incubation in Rec4U buffer (20 mM Tris·HCl, pH 7.6, 4 mM magnesium acetate, 0.5 M NH₄Cl, 6 M urea) for 2-4 h at 40 °C. The reaction was quenched by adding one volume of acetic acid and cooling to 0 °C. Protein was precipitated by addition of five volumes of acetone and overnight storage at 20 °C, and dissolved in a 3:1 mixture of 0.1% trifluoroacetic acid:REC20U buffer (20 mM Tris·HCl, pH 7.6, 20 mM magnesium acetate, 0.5 M NH₄Cl, 6 M urea, 10 mM -mercaptoethanol) prior to RP-HPLC fractionation.

Reconstitutions were carried out by combining rRNA, [³H]DNP-L23, TP50-L23, and placental RNase inhibitor protein (4 units/A₂₆₀ equivalent) as described for the reconstitution of 50 S subunits containing puromycin-L23 in place of L23 (6). The molar ratio of protein to RNA was 1.8. Proteins dissolved in REC4U were dialyzed versus REC4 (lacking urea) before being used in reconstitution. Reconstituted subunits were stored at 70 °C in 50 mM Tris·HCl, pH 7.6, 10 mM MgCl₂, 50 mM KCl.

In a typical hydrolysis, 25 µg of [³H]DNP-L23 in 0.1% trifluoroacetic acid was added to an ampoule containing 40 µg each of N-DNP-Lys and N-DNP-Met and lyophilized. The sample was dissolved in 400 µl of 6 N HCl, purged with nitrogen gas, sealed, and heated for 10-16 h at 105 °C. Following transfer and lyophilization, the sample was dissolved in 10 µl methanol followed by 1 ml of 0.1% trifluoroacetic acid for analysis by RP-HPLC. [³H]DNP-L23 derived from reconstituted 50 S subunits was partially purified by RP-HPLC on a C18 column prior to acid hydrolysis. This procedure also removed contaminating sucrose and nucleic acid that interfered with the analysis of hydrolysis products.

Two samples of reconstituted subunits were used. One sample was prepared using protein that contained 1.3 DNP/L23, and the resulting 50 S particles contained 0.65 DNP/50 S. A second sample was prepared with protein that contained 0.9 DNP/L23, yielding subunits that included 0.4 DNP/50 S. In one set of experiments, 14-16 pmol (0.35-0.40 A₂₆₀ units) of reconstituted subunits were incubated with 2-4 binding equivalents (relative to the DNP content of the particles) of anti-DNP IgG, based on two equivalents of DNP binding capacity per IgG measured as described (32). The total volume was 20-35 µl of 25-40 mM Tris·HCl, pH 7.5, 150 mM NH₄Cl, 10 mM MgCl₂. Samples were incubated at 37 °C for 5-15 min and then on ice for 12-16 h. They were then fractionated by size exclusion HPLC using either a 7.5 × 300-mm Beckman Spherogel TSK 3000 SW column or a 7.8 × 300-mm Supelco Progel TSK 3000 SW XL column that had been equilibrated with 10 mM Tris·HCl, pH 7.5, 150 mM NH₄Cl, 10 mM MgCl₂ (10/150/10 buffer) as described (33).

In a second set of experiments, 15-16 pmol (0.37-0.40 A₂₆₀ units) of reconstituted 50 S subunits were incubated first with 60-120 pmol (4-8 molar equivalents) of HPLC-purified protein L7/L12 in 10-25 µl of 40-50 mM Tris·HCl, pH 7.5, 50 mM KCl, 10 mM MgCl₂ at 37 °C for 15 min. Then 2-4 binding equivalents of anti-DNP IgG were added, and the buffer was adjusted with NH₄Cl to raise the monovalent cation level to 150 mM. The samples were incubated at 37 °C for an additional 5-15 min followed by 12-16 h on ice, and fractionated as described above.

Reconstituted 50 S subunits were also studied in 70 S ribosomes. About 8 pmol (0.2 A₂₆₀ unit) of reconstituted 50 S particles containing DNP-modified L23 were mixed with 16 pmol (0.24 A₂₆₀ unit) of 30 S subunits in 10 ml of 10/150/10 buffer and incubated at 37E C for 15 min. Then 4 binding equivalents of anti-DNP IgG in 2 µl of 25/150/10 buffer were added, and the mixtures were incubated for 15 min at 37 °C followed by 16 h on ice. Immune complexes were fractionated at 0 °C as above, using 10/150/10 buffer. Alternatively, 8 pmol (0.2 A₂₆₀ unit) of reconstituted 50 S particles containing DNP-modified L23 were incubated with 4 binding equivalents of anti-DNP IgG in 10 µl of 10/150/10 buffer for 15 min at 37 °C and then on ice for 12 h. Then 16 pmol (0.24 A₂₆₀ unit) of 30S subunits in 5 µl of 25/150/10 buffer were added, and samples were incubated for 10 min at 37 °C, followed by HPLC fractionation as described above.

Size-exclusion HPLC fractions containing ribosomes and immune complexes were immediately adsorbed to thin carbon films and negatively contrasted with 0.7% (w/v) uranyl acetate using the double carbon technique as described previously (18, 25). Electron micrographs were obtained and evaluated as described (25). Nomenclature is from Lake (18) .

RESULTS

Confidence in our interpretation of the immunoelectron microscopy depends on the DNP derivatization of highly purified L23, i.e. sites of antibody attachment must reflect binding to derivatized L23 only, rather than to L23 and other proteins. We purified L23 through the use of multiple steps of HPLC, using both C18 and C8 columns, as described in Fig. 1. The final material was homogeneous on both RP-HPLC and SDS-PAGE analysis (Fig. 2) and had the A₂₈₀/A₂₁₅ ratio expected for a protein containing one Trp and one Tyr (0.047). Protein L22, containing only one Tyr and no Trp, has a much lower value: 0.007, and L29, lacking both Tyr and Trp has a value of <0.001 (see Ref. 29). Simultaneously satisfying all of these criteria is important since two proteins of the 70 S ribosome (S14 and L24) have molecular weights similar to L23 and are not well resolved from it by SDS-PAGE, but elute far from L23 on RP-HPLC (34). Finally, quantitative amino acid sequence analysis gave results consistent with L23 and ruled out more than trace contamination with the two proteins, L22 and L29, that elute close to L23 on RP-HPLC (Table I). We estimate that the L23 preparation derivatized with DNFB was 98% pure.

Table I. N-terminal sequencing of purified L23 Shown are N-terminal sequences for L22, L23, and L29 (67) along with the yields of the three relevant amino acids obtained in each cycle. Sequence was determined for purified L23 (Fig. 2) by N-terminal Edman degradation in the Microsequencing Facility of the Wistar Institute. L22   M     E     T   I     A     K     H     R    190.1 1.9 0.6 1.4 5.6 4.6 0.6 3.1 L23   M     I     R     E     E     R     L     L    190.1 108.7 108.5 165.9 207.2 135.4 156.0 183.2 L29   M     K     A     E     L     R     E     K    190.1 0.0 2.2 165.9 4.0 135.4 7.8 0.0

Modification conditions were chosen to afford a stoichiometry of approximately 1 DNP/L23, based on the earlier observation of Olah et al. (28) that, at least in 30 S reconstitution, less modified proteins are selected over more heavily modified proteins. Prior to use in reconstitution, residual unreacted L23 was removed by RP-HPLC fractionation (Fig. 3), in order that it not compete with [³H]DNP-L23 for reconstitution into 50 S subunits.

DNFB was originally introduced by Sanger (35, 36) as a reagent to specifically modify the -amino group of a protein, Met in the case of L23. The major competing reaction is with the -amino group of internal lysines. N-DNP-Met and N-DNP-Lys are readily separable by RP-HPLC, allowing quantification of modification sites following acid hydrolysis of DNP-L23. The N-DNP-Met/N-DNP-Lys ratio (Fig. 4) was determined both for DNP-L23 prepared by reaction of DNFB with L23 and for DNP-L23 partially purified from reconstituted 50 S subunits (see below). In the modification of L23 we find a 1:1 ratio, i.e. the 13 Lys residues in L23 have a cumulative reactivity equal to that of the single N-terminal Met. This ratio shifts to about 3:2 in the L23 incorporated by 50 S subunits,

50 S reconstitutions were performed by combining [³H]DNP-L23, TP50-L23, and rRNA. When a sample containing 1.3 DNP/L23 was employed the resulting particles contained 0.65 ± 0.02 DNP/50 S, as measured following sucrose density gradient purification of reconstituted 50 S. This figure did not change following purification on a second sucrose density gradient and on a size-exclusion column, thus demonstrating that DNP-L23 is stably incorporated into the 50 S subunit. Unmodified L23 competes efficiently with DNP-L23 in reconstitution; when each was present at an equal concentration, sucrose gradient analysis showed that incorporation of [³H]DNP-L23 was reduced by at least half.

Ribosomal subunits containing DNP-L23 were reconstituted with a preparation of TP50 that was depleted for protein L7/L12. Electron micrographs of this preparation showed very few particles that displayed the stalk that is characteristic of protein L7/L12 (33), but the subunit images were otherwise typical of 50 S ribosomal subunits. These subunits were incubated with purified protein L7/L12 and examined by electron microscopy. Roughly half of the particles showed a stalk; a similar proportion of stalks was seen in preparations of native 50 S subunits and in subunits that had been treated with NH₄Cl and ethanol to remove the stalk and then incubated with purified L7/L12 to restore it. We conclude that by the criterion of electron microscopy these reconstituted subunits have the characteristics of native particles.

Reconstituted subunits containing DNP-L23 were incubated with a small excess of anti-DNP IgG; uncomplexed antibodies were removed by size exclusion HPLC and the fraction containing immune complexes and 50 S subunits was prepared for electron microscopy. Two different preparations were studied in these experiments. In the first case, a total of 125 micrographs showing 7.8 × 10³ subunits were analyzed, and 200 immune complexes were identified and interpreted. In the second case we analyzed 43 micrographs showing about 4.3 × 10³ particles and identified and interpreted 154 immune complexes. The location of the DNP hapten, and thus the placement of the modified protein in the three dimensional structure of the ribosomal subunit, was identified from the apparent point of contact of the antibody molecule with the subunit as seen in each of its characteristic two-dimensional projections, each of which can be seen in two orientations that are mirror images of each other (18). The results of these observations are summarized in Table II.

Table II. Quantitation of antibody-subunit contact points a 125 micrographs (7.8 × 103 subunits) analyzed; 9 contacts in antibody-linked dimers. b 178 micrographs (1.4 × 104 subunits) analyzed; 17 contacts in antibody-linked dimers. c 43 micrographs (4.3 × 103 subunits) analyzed; 8 contacts in antibody-linked dimers. d 110 micrographs (1.1 × 103 subunits) analyzed; 20 contacts in antibody-linked dimers. e Numbers in parentheses represent percentage of total.

Antibodies were observed to be bound at either of two major sites: beside the central protuberance and, with slightly higher frequency, at the base of the particle. The gallery of complexes shown in Fig. 5 illustrates the predominant observations upon which this conclusion is based. Subunits in rows A and B are in the quasisymmetric projection. In row A antibody contact is near the central protuberance with the contacting Fab arm at least partially obscured by the subunit body. In row B antibody contact is seen at the base of the subunit, slightly removed from the vertical axis of near symmetry and with the antibody Fab arm partially obscured by the subunit body. Rows C and D show subunits in the asymmetric projection. In row C antibody contact is in the region of the central protuberance, while in row D contact is on the surface opposite the central protuberance. Subunits in row E are shown in the quasisymmetric projection; two antibodies are bound to a single subunit, one antibody at each site identified above.

The subunits shown in Fig. 5 were reconstituted from protein mixtures that lacked L7/L12, the protein that generates the stalk of 50 S subunits, and no stalks were seen on subunits in these micrographs. The absence of the stalk makes antibody identification much less ambiguous, but the stalk is a major asymmetric element in the quasisymmetric projection of the 50 S subunit. To establish the side of the line of near symmetry at which antibody is bound, it was necessary to restore the stalk to these subunits. Reconstituted subunits containing DNP-L23 were first incubated with a 2-fold excess of protein L7/L12 and then with anti-DNP IgG. Unincorporated L7/L12 and uncomplexed antibodies were removed by size exclusion HPLC, and the fraction containing 50 S subunits and immune complexes was prepared for electron microscopy. The field of Fig. 6A shows that in at least half of the subunits the characteristic stalk is present. From these micrographs we have identified and analyzed 94 antibody-subunit complexes in which a stalk is also clearly seen on subunits in the quasisymmetric projection; panels B-D of Fig. 6 illustrate these results. In panel B antibody contact is near or beside the central protuberance and on the side opposite the stalk. In panel C antibody contact is at the base of the subunit and also on the side opposite the L7/L12 stalk, while in panel D two antibodies are bound, one at each of the sites identified above.

In our micrographs we also observed several dimeric complexes in which a pair of subunits is linked by a single bivalent IgG molecule. Subunits were seen in each of the orientations described above, and antibody contact sites were consistent with the observations made on monomeric IgG complexes. Examples are shown in Fig. 7. Antibody contact to each subunit is seen near the central protuberance (frames 1 and 2) or at the subunit base (frame 3), or one subunit is contacted near the central protuberance while the second subunit is bound at the base (frames 4-6). Those subunits in the quasisymmetric projection that show a stalk (in frames 1 and 5 the subunit on the right; in frame 6 the upper subunit) are contacted by antibody on the side opposite the stalk. The lower subunit in frame 6 also shows a stalk, but it is in the asymmetric projection so it is not possible to determine on which side of the central protuberance antibody contact occurs.

The 50 S particles that had been reconstituted with DNP-L23 could associate with 30 S subunits to form 70 S ribosomes that bound anti-DNP IgG. Examples of such complexes are shown in frames 7-9 of Fig. 7; the ribosomes are oriented with the 30 S subunit lying atop the 50 S particle. In frames 7 and 8 antibody contact is seen at the base of the larger subunit, well removed from the smaller (head) segment of the small subunit. Such complexes were seen with a frequency similar to that observed with 50 S subunits. Frame 9 shows an example in which the site of antibody contact is at or very near the plane of subunit interaction and on the same end as the 30 S subunit head, indicating contact in the peptidyltransferase region. This type of complex was very rare. The result was expected; the 30 S subunit should block contact at the transferase region, and in addition any bound antibody could easily be obscured by the ribosome. Nevertheless, the sites we observe with 70 S ribosomes are compatible with the observations made with 50 S subunits. We conclude that the DNP moiety of DNP-L23 can be placed at two apparently distant sites. Fig. 8 illustrates our observations using the Frank model of the 50 S subunit, which was reconstructed from electron micrographs of frozen unstained particles (22).

DISCUSSION

The validity of the conclusions in Fig. 8 depends entirely on the successful replacement of native L23 by DNP-L23 in the reconstituted 50 S subunits. Previously, we demonstrated the validity of this approach using DNP-modified proteins of the 30 S subunit (25-28). In accord with the earlier results, incorporated DNP-L23 appears to bind within the 50 S subunit in the same or similar manner as native L23, as evidenced by the ability of unmodified L23 to efficiently compete with [³H]DNP-L23 uptake. This result provides strong evidence that the localization of DNP-L23 within reconstituted 50 S subunits by immune electron microscopy faithfully reflects the location of unmodified L23, at the resolution achievable by this technique.

The results we present here indicate that, as shown in Fig. 8, ribosomal protein L23 spans the 50 S subunit, from a site at the subunit interface and at or near the peptidyltransferase center (37) to a second site, at the base of the 50 S particle and indistinguishable from the exit site (14) at which the growing peptide chain emerges from the ribosome. There are several potential explanations for this unexpected observation, including: (i) more than one physical site in the ribosome for protein L23, (ii) a very elongated, asymmetric conformation for L23, and (iii) inaccuracy or inappropriateness of the 50 S models commonly used. We believe that the combination of a somewhat asymmetric conformation of L23 and a porous and indented 50 S subunit surface adequately explains both our results and a large body of apparently contradictory data that place protein L23 in two distant parts of the subunit.

Early work on the placement of ribosomal proteins by IEM often resulted in the identification of multiple sites for a single protein, many of which were incorrect (38). These errors were due to the use of impure ribosomal proteins as immunogens. The resulting antisera interacted with several proteins, and, unfortunately, the results were interpreted as if the antibodies were monospecific. The approach used here excludes this kind of error. Protein L23 was HPLC-purified prior to modification, and any possible contamination (<2%) is far less than that necessary to constitute one of the two binding sites we observe. It is equally improbable that a compact, globular protein can occupy two different sites at two separated positions in the 50 S particle. Protein L23 exists in one copy per 50 S particle and it enters the large subunit early in its assembly (4). Subunit biogenesis is an ordered process that involves specific interactions of the proteins with both the RNA and other ribosomal proteins. L23 is a primary binding protein that binds and protects from RNase action a terminal loop and part of an internal loop in 23S RNA domain III (3); this binding motif is phylogenetically conserved in large subunit rRNA and is required for L23 binding to eubacterial and chloroplast RNA and for the binding of its eukaryotic homologue, L25 (39, 40). Inactivation of the yeast genes for either the mitochondrial or cytoplasmic ribosome homologues of L23 is lethal (41, 42), suggesting a central role in the ribosome. It is difficult to see how the protein could participate in an alternative set of highly specific but mutually exclusive interactions with the RNA, and we are unaware of any other ribosomal protein with such properties. We conclude that each of the two sites we identify represents a valid localization of a part of a single molecule of L23.

There is evidence to support the location of parts of a possibly asymmetric protein L23 at each site we identify here. Photoaffinity labeling of the 50 S subunit by puromycin (7) or azidopuromycin (8) results in their predominant incorporation into L23. Puromycin is an acceptor in the peptidyltransferase reaction and functionally defines the A site of the transferase center. Puromycin-modified L23 has been incorporated into reconstituted 50 S subunits, with a proportional decrease in mRNA-dependent aminoacyl-tRNA binding activity but no specific effect on the peptidyltransferase reaction (6). Tetracycline interacts directly with the central loop of domain V, the region of 23 S rRNA most strongly implicated in peptidyltransferase activity (43, 44), and it also specifically stimulates incorporation of puromycin into protein L23 (7). Streptomycin affects the elongation step of protein synthesis primarily by interfering in ternary complex binding. Streptomycin derivatives that photoaffinity-label large subunits in 70 S ribosomes are incorporated into L23 and other subunit interface proteins (45). As concluded earlier (6), these antibiotic studies indicate that L23 is near the transferase center and probably a structural part of the A-site, but not a catalytic component of the transferase. Puromycin and azidopuromycin, incorporated primarily into L23 by photoaffinity labeling, have been localized by IEM at a site beside the central protuberance (10, 11), helping to define the peptidyltransferase region on the 50 S subunit and placing L23 near the transferase center. Affinity labeling with a bromoacetyl derivative of puromycin also places the antibiotic at the transferase center, on the subunit shoulder, and results in the modification of proteins L1, L2, L23, and L27 (46). Cross-linking of L23 to 30 S proteins (47) puts it at the subunit interface, and cross-linking to 50 S proteins L5, L15, L16, L18, and L27, all of which have been localized at or near the central protuberance by IEM and linked to transferase activity by functional studies (37), further indicates a site at or near the transferase center.

Early IEM from the Stöffler laboratory (48) used polyclonal antibodies directed against L23 and placed it in the transferase region; additionally, anti-L23 Fab fragments were found to inhibit subunit association, suggesting an interface site (49). However, uncertainty about antibody specificity led the Berlin workers to question their results. Well controlled IEM placed both E. coli (13) and Bacillus subtilis (50) protein L23 at the subunit base, at the position of the exit site (as marked by an antibody initially thought to be specific for L23 but later designated LY; Ref. 51). Placement of L23 at the exit site in the model of Walleczek et al. (52) was based on IEM; it is supported by cross-linking that places the N terminus of L23 near L29 (15) and by cross-linking of a segment of 23 S rRNA to both L23 (53) and the N terminus of a 30-33 residue long emerging peptide (54). However, identification of an epitope at the exit site does not rule out the possibility that other elements of the protein, not recognized by the antibody preparation, exist at a separate position in the particle.

Several results support the existence of two separate sites for distinct elements of L23 on different surfaces of the subunit. Cross-linking of the protein to nucleotides near the 5 end of the 23 S rRNA (53) in conjunction with identification of the primary L23 binding site in domain III (3) indicates that the protein interacts simultaneously with separate elements of RNA tertiary structure. Mueller et al. (55) consider "circumstantial evidence" that L23 lies along the path of the nascent peptide: a 25-residue-long peptide also cross-links to RNA domain III while a longer peptide cross-links at nucleotide 91 (54), both near the L23 sites. A C-terminal region of the yeast homologue of L23 interacts with rRNA in a crucial assembly step that follows RNA binding (56), indicating a role for the protein in forming or maintaining RNA tertiary structure. In work from our laboratories, IEM of 50 S particles that were photoaffinity-labeled with both puromycin and azidopuromycin showed a secondary site at the base of the particle that accounted for up to 25% of the complexes seen (10, 11). It was unclear if the second site was a result of nonspecific binding, a genuine placement of puromycin-L23, or unrelated to L23.

In part to accommodate the apparently conflicting data on L23 localization, Nagano et al. (57) suggested that L23 is a dumbbell-shaped protein in which the N terminus and C terminus form independent domains that are separated by 90 ± 30 Å. Protein L9 provides an example of such asymmetry in the ribosome. The crystal structure (58) shows two RNA-binding globular domains linked by an exposed -helix, yielding a protein that is 82 Å long. IEM places an epitope of L9 near the L1 shoulder, but antibody binding most strongly affects activities at the translocational domain, between the central protuberance and the base of the L7/L12 stalk (59).

Asymmetry of protein L23 may partially explain our results, but we also believe that the common (visually derived) models of the 50 S subunit (38) do not accurately describe the subunit and the channel (60) or tunnel (22, 23, 61) that is traversed by the growing peptide chain. Approximately 25-40 amino acids of the nascent peptide are protected by the ribosome from proteinase K, and slightly greater lengths are protected from access to IgG or Fab fragments (62). Hardesty et al. (62) estimate that this peptide length (considered to be -helical) indicates a transferase-exit site separation of 65-105 Å, significantly less than the 150 Å estimate (14) which was based on the separation of these sites in the common 50 S models. Protein L23 has a length of 100 amino acids (63), and the isolated protein has an axial ratio of 2.8 ± 0.9 (64). It is unlikely that it could span the full distance in the usual models, but the two sites appear to be much closer together (about 80 Å) in structures reconstructed from images of unstained frozen ribosomes (22, 23, 65). Reconstructions show several pockets or cavities in the 50 S subunit, as well as the tunnel that traverses the particle and links the transferase and exit sites (see Fig. 8). Images reconstructed from micrographs of two dimensional arrays of subunits also show a tunnel (66). As shown in Fig. 8, one of our L23 sites is in the the "interface canyon" (65) and at the peptidyltransferase end of the tunnel, while the second site includes indentations on the back of the particle and an exit from the tunnel. The relatively porous structure of the subunit at each site could provide at least partial access to an Fab arm and thus allow antibody to approach different elements of L23 from two separate directions. From a different perspective, our results represent an independent confirmation of the indented and porous structure of the ribosome, as it is seen in cryo-EM reconstructions. It may be necessary to re-interpret much of the protein localization literature in this light.

L23 can thus be described as a protein of the subunit interior with segments surfacing near the tunnel entrance and its exit. Analysis of DNP-L23 in reconstituted subunits indicates that the Met derivative is slightly preferred in reconstitution to the average Lys derivative and that reconstitution may select against L23 molecules in which some specific Lys residues have been dinitrophenylated. In the present work, no attempt was made to quantify the extent of modification at specific lysine residues. We emphasize that whereas N-terminal derivatization occurs at a well defined position within a protein, derivatization of -amino groups can occur at several of the 13 lysines within L23. This point is important for interpretation of the immune electron microscopy results. Since about 60% of the DNP incorporated by L23 is at the N-terminal methionine, which has been cross-linked to protein L29 (15) and thus placed at the subunit base (52), and a similar percentage of antibody complexes are seen at the subunit base, it is likely that DNP-Met generates these complexes. We see no evidence for DNFB modification at positions other than the N terminus or the -amino group of internal lysines for any of the proteins we have studied. This may be due to the instability of other DNP derivatives toward nucleophilic displacement by thiols (8-mercaptoethanol is a component of buffer REC20U), as discussed earlier (26). Modification of one or more internal lysine residues would then generate the complexes we see near the peptidyltransferase center. L23 would thus lie parallel to or along the 50 S tunnel with its N terminus at the exit site and the more flexible C terminus near the transferase center.