Receptor binding kinetics of human IL-3 variants with altered proliferative activity.

The binding kinetics of native IL-3 and a set of truncated IL-3 variants to the alpha subunit of the IL-3 receptor (IL-3Ralpha) were studied using surface plasmon resonance. These variants, with amino acid substitutions at residues, 22, 42, 43, 45, 46, 113, or 116, have previously been identified to have altered capacity to stimulate cell proliferation compared to native IL-3(1-133). In this study, variants E43N and F113Y exhibited >100-fold slower association rates than IL-3(15-125) consistent with residues 43 and 113 being essential for the binding of IL-3 to the IL-3Ralpha. Variants G42A, G42D, Q45V, D46S, K116V, and K116W exhibited increased association rates (up to 15-fold relative to IL-3(15-125)) and decreased dissociation rates (up to 7-fold). The results demonstrate that both the association and dissociation rates for the binding of IL-3 to the IL-3Ralpha are altered by truncation and by amino acid substitution at individual sites. Intracellular signaling studies using K116W and E43N demonstrate that differences in the IL-3alpha binding characteristics are reflected in magnitude and kinetics of STAT5 phosphorylation.

Modulation of the binding affinity of myelopoietins for the interleukin-3 receptor by the granulocyte colony-stimulating factor receptor agonist.

Myelopoietins (MPOs) are a family of recombinant chimeric proteins that are both interleukin-3 (IL-3) receptor and granulocyte colony-stimulating factor (G-CSF) receptor agonists. In this study, MPO molecules containing one of three different IL-3 receptor agonists linked with a common G-CSF receptor agonist have been examined for their IL-3 receptor binding characteristics. Binding to the alpha-subunit of the IL-3 receptor revealed that the affinity of the MPO molecules was 1.7-3.4-fold less potent than those of their individual cognate IL-3 receptor agonists. The affinity decrease was reflected in the MPO chimeras having approximately 2-fold slower dissociation rates and 2.7-5.5-fold slower association rates than the corresponding specific IL-3 receptor agonists alone. The affinity of binding of the MPO molecules to the heteromultimeric alphabeta IL-3 receptor expressed on TF-1 cells was either 3-, 10-, or 42-fold less potent than that of the individual cognate IL-3 receptor agonist. Biophysical data from nuclear magnetic resonance, near-UV circular dichroism, dynamic light scattering, analytical ultracentrifugation, and size exclusion chromatography experiments determined that there were significant tertiary structural differences between the MPO molecules. These structural differences suggested that the IL-3 and G-CSF receptor agonist domains within the MPO chimera may perturb one another to varying degrees. Thus, the differential modulation of affinity observed in IL-3 receptor binding may be a direct result of the magnitude of these interdomain interactions.

Circular permutation of the granulocyte colony-stimulating factor receptor agonist domain of myelopoietin.

Myelopoietins (MPOs) are a family of engineered dual interleukin-3 (IL-3) and granulocyte colony-stimulating factor (G-CSF) receptor agonists that are superior in comparison to the single agonists in their ability to promote the growth and maturation of hematopoietic cells of the myeloid lineage. A series of MPO molecules were created which incorporated circularly permuted G-CSF (cpG-CSF) sequences with an IL-3 receptor (IL-3R) agonist moiety attached at locations that correspond to the loops that connect the helices of the G-CSF four-helix bundle structure. The cpG-CSF linkage sites (using the original sequence numbering) were residue 39, which is at the beginning of the first loop connecting helices 1 and 2; residue 97, which is in the turn connecting helices 2 and 3; and residues 126, 133, and 142, which are at the beginning, middle, and end, respectively, of the loop connecting helices 3 and 4. The N- and C-terminal helices of each cpG-CSF domain were constrained, either by direct linkage of the termini (L0) or by replacement of the amino-terminal 10-residue segment with a seven-residue linker composed of SGGSGGS (L1). All of the MPO molecules stimulated the proliferation of both IL-3-dependent (EC50 = 13-95 pM) and G-CSF-dependent (EC50 = 35-710 pM) cell lines. MPOs with the IL-3R agonist domain linked to cpG-CSFs in the first (residue 39) or second (residue 133) long overhand loops were found by CD spectroscopy to have helical contents similar to that expected for a protein comprised of two linked four-helix bundles. The MPOs retained the ability to bind to the IL-3R with affinities similar to that of the parental MPO. Using both a cell surface competitive binding assay and surface plasmon resonance detection of binding kinetics, the MPOs were found to bind to the G-CSF receptor with low nanomolar affinities, similar to that of G-CSF(S17). In a study of isolated cpG-CSF domains [Feng, Y., et al. (1999) Biochemistry 38, 4553-4563], domains with the L1 linker had lower G-CSF receptor-mediated proliferative activities and conformational stabilities than those which had the L0 linker. A similar trend was found for the MPOs in which the G-CSFR agonist activity is mostly a property of the cpG-CSF domain. Important exceptions were found in which the linkage to the IL-3R agonist domain either restored (e.g., attachment at residue 142) or further decreased (linkage at residue 39) the G-CSFR-mediated proliferative activity. MPO in which the IL-3R agonist domain is attached to the cpG-CSF(L1)[133/132] domain was shown to be more potent than the coaddition of the IL-3R agonist and G-CSF in stimulating the production of CFU-GM colonies in a human bone marrow-derived CD34+ colony-forming unit assay. Several MPOs also had decreased proinflammatory activity in a leukotriene C4 release assay using N-formyl-Met-Leu-Phe-primed human monocytes. It was found that circular permutation of the G-CSF domain can alter the ratio of G-CSFR:IL-3R agonist activities, demonstrating that it is a useful tool in engineering chimeric proteins with therapeutic potential.

Use of combinatorial mutagenesis to select for multiply substituted human interleukin-3 variants with improved pharmacologic properties.

A combinatorial mutagenesis strategy was used to create a collection of nearly 500 variants of human interleukin 3 (IL-3), each with four to nine amino acid substitutions clustered within four linear, nonoverlapping regions of the polypeptide. The variants were secreted into the periplasm of Escherichia coli and supernatants were assayed for IL-3 receptor-dependent cell proliferation activity. Sixteen percent of the variants, containing "region-restricted" substitutions, retained substantial proliferative activity through two rounds of screening. A subset of these was combined to yield variants with substitutions distributed through approximately half of the polypeptide. With one exception, "half-substituted" variants exhibited proliferative activity within 3.5-fold of native IL-3. A subset of the "half-substituted" variants was combined to yield "fully substituted" IL-3 variants having 27 or more substitutions. The combination of the substitutions resulted in a set of polypeptides, some of which exhibit increased proliferative activity relative to native IL-3. The elevated hematopoietic potency was confirmed in a methylcellulose colony-forming unit assay using freshly isolated human bone marrow cells. A subset of the multiply substituted proteins exhibited only a modest increase in inflammatory mediator (leukotriene C4) release. The molecules also exhibited 40- to 100-fold greater affinity for the alpha subunit of the IL-3 receptor and demonstrated a 10-fold faster association rate with the alpha-receptor subunit. The multiply substituted IL-3 variants described in this study provide a unique collection of molecules from which candidates for clinical evaluation may be defined and selected.

The receptor binding site of human interleukin-3 defined by mutagenesis and molecular modeling.

Interleukin-3 (IL-3) is a member of the cytokine superfamily that promotes multi-potential hematopoietic cell growth by interacting with a cell surface receptor composed of alpha and beta chains. The newly available three-dimensional structure of a variant of human (h) IL-3 allowed us to evaluate new and existing mutagenesis data and to rationally interpret the structure-function relationship of hIL-3 on a structural basis. The amino acid residues that were identified to be important for hIL-3 activity are grouped into two classes. The first class consists of largely hydrophobic residues required for the structural integrity of the protein, including the residues in IL-3 that are largely conserved among 10 mammalian species. These residues form the core of a scaffold for the second class of more rapidly diverging solvent-exposed residues, likely to be required for interaction with the receptor. Ten important and solvent-exposed residues, Asp21, Gly42, Glu43, Gln45, Asp46, Met49, Arg94, Pro96, Phe113, and Lys116, map to one side of the protein and form a putative binding site for the alpha subunit of the receptor. A model of the IL-3.IL-3 receptor complex based on the human growth hormone (hGH).hGH soluble receptor complex structure suggests that the interface between IL-3 and the IL-3 receptor alpha subunit consists of a cluster of hydrophobic residues flanked by electrostatic interactions. Although the IL-3/IL-3 receptor beta subunit interface cannot be uniquely located due to the lack of sufficient experimental data, several residues of the beta subunit that may interact with Glu22 of IL-3 are proposed. The role of these residues can be tested in future mutagenesis studies to define the interaction between IL-3 and IL-3 receptor beta subunit.

Three-dimensional solution structure and backbone dynamics of a variant of human interleukin-3.

The three-dimensional structure and backbone dynamics of a truncated and multiply substituted recombinant human interleukin-3 (IL-3) variant (SC-65369) have been determined from multidimensional heteronuclear nuclear magnetic resonance spectroscopic data. Sequential application of distance geometry and restrained molecular dynamics calculations produced a family of 25 convergent structures which satisfy a total of 1812 experimental constraints (1659 proton-proton NOEs, 75 backbone dihedral angle constraints, and 39 pairs of hydrogen bond constraints) with an average root-mean-square deviation from the mean coordinate positions of 0.88(+/- 0.15) angstroms and 1.37(+/- 0.13) angstroms for the backbone and all heavy atoms, respectively, of all residues except 28 to 39. The structure is a left-handed four-helix bundle (comprised of helices A through D) with two long overhand loops (designated as loops AB and CD). Loop AB contains a short fifth helix (helix A') which is closely packed against helix D in an approximately parallel fashion and which has multiple contacts with loop CD. The overall molecular tumbling time (6.5 ns) determined from the 15N relaxation data was consistent with a monomeric protein under the conditions of the experiment (1 mM protein, pH 4.6, 30 degrees C). The 15N relaxation data indicate that the helical regions of SC-65369 are quite rigid, while portions of loop AB, loop CD, and the C terminus undergo significant internal motions. Among the structurally related four-helical bundle cytokines, the structure of SC-65369 is most similar to those of granulocyte-macrophage colony stimulating factor (GM-CSF) and the single structural domain of interleukin-5 (IL-5), all of which share a common receptor subunit required for signal transduction and activation of their hematopoietic target cells. Indeed, the C(alpha) atoms in the four-helix core of these three proteins can be superimposed to 1.71 angstroms (SC-65369 and GM-CSF, 62 C(alpha) atoms) and 1.96 angstroms (SC-65369 and IL-5 single structural domain, 58 C(alpha) atoms), respectively. When the structures of the IL-3 variant, GM-CSF, and IL-5 were aligned, the conserved and conservatively substituted residues were found to be hydrophobic and buried, with the single exception of Glu-22 (IL-3 numbering), which is strictly conserved but nonetheless fully exposed to solvent. The most remarkable differences between the SC-65369 structure and that of GM-CSF occur in loop AB. This loop in GM-CSF crosses over the top of helix D and passes underneath loop CD on its way to helix B. In contrast, loop AB of SC-65369 passes in front of helix D, similar to the first crossover loop in human growth hormone and granulocyte colony-stimulating factor. In addition, helix A', which is interdigitated into the helical bundle in a manner similar to the helices in the CD loop of interferon-beta and interferon-gamma, exists in a region where short stretches of beta-structure are found at analogous positions in GM-CSF and IL-5. These differences suggest that the structural elements within this region may be important for recognition by their cognate receptors.

Saturation mutagenesis of human interleukin-3.

A deletion variant of human interleukin-3, hIL-3(15-125), was produced in the periplasmic space of Escherichia coli and had full activity in an AML193.1.3 cell proliferation assay. Libraries of random single-amino acid substitutions were constructed at each of 105 positions in the gene for hIL-3(15-125). Approximately eight single-site substitutions at each position were produced in osmotic shock fractions and screened for activity. 15 mutants were found with bioactivity of 5-26-fold greater than that of native hIL-3. The majority of amino acids in hIL-3(15-125) could be substituted without substantial loss of activity. Substitution of residues predicted to be in the hydrophobic core of the protein often resulted in reduced activity and/or low accumulation levels. Only five residues predicted to be on the surface of the protein were intolerant of substitution and hence are candidates for sites of interaction with the receptor. We therefore propose that the majority of residues in hIL-3 serve a structural role and permit the display of a few key residues in the correct configuration for recognition by the receptor.

1H, 13C, and 15N NMR resonance assignments, secondary structure, and backbone topology of a variant of human interleukin-3.

Interleukin-3 (IL-3) is a cytokine which stimulates the proliferation and differentiation of hematopoietic progenitors into multiple cell lineages. The 1H, 15N, and 13C NMR resonances of a recombinant human IL-3 variant (SC-65369) have been assigned using two- and three-dimensional NMR techniques on uniformly 13C/15N-enriched protein. Five helical segments (residues 16-26, 42-50, 55-65, 73-82, and 104-120) and three reverse turns (residues 51-54, 68-71, and 87-90) were identified from the pattern of sequential NOE connectivities, NH(i)-C alpha H(i) scalar coupling constants (3JNH alpha), amide hydrogen exchange data, and the deviation of 13C alpha, 13C beta, 13CO, and C alpha H chemical shifts from random-coil values. Long-range NOEs indicate that the global folding pattern of human IL-3 is a four-helical bundle with an up-up-down-down arrangement of helices that is similar to that of other members of the cytokine family, such as granulocyte-macrophage colony stimulating factor (GM-CSF). A fifth short helix (helix A', residues 42-50) is located in the loop connecting the first and second helices. The absence of helix A' in the corresponding structures of GM-CSF and interleukin-5 suggests that it may be important for recognition of IL-3 by its receptor. The existence of at least two forms of the protein that differ in local conformation was implied from the observation of a limited set of doubled resonances in which each doublet partner had a similar pattern of short-, medium-, and long-range NOEs. The majority of the doubled resonances were close in sequence or space to a proline-rich sequence, which suggested that the different conformational forms of SC-65369 may be caused by slow cis-trans isomerization of proline peptide bonds.

Effects of signal peptide changes on the secretion of bovine somatotropin (bST) from Escherichia coli.

Bovine somatotropin (bST) was secreted from Escherichia coli at moderate levels of 1-2 micrograms/ml/OD using expression vectors in which the bST gene was fused to the lamB secretion signal. To study the secretion properties of bST in E.coli further, two approaches for modifying the secretion signal were employed. In the first case, fusion proteins were constructed with six alternative bacterial secretion signals: three from E.coli proteins (HisJ, MalE and OmpA), two from bacteriophage proteins (M13 coat protein and PA-2 Lc) and one from the chitinase A protein of Serratia marcescens. The results, as monitored by Western blot analysis of both total cell protein and the periplasmic fraction, showed that these changes in the secretion signal did not significantly affect the secretion properties of bST. In the second approach, a library of random mutations was created in the lamB secretion signal and 200 independent clones were screened. The level of secreted bST was determined by growing individual clones in duplicate in microtiter wells, inducing protein expression and measuring the bST released by osmotic shock using a particle concentration fluorescent immunoassay. The secretion properties of several novel variants in the LamB signal peptide are presented.

Secretion of active bovine somatotropin in Escherichia coli.

We have expressed a chimeric protein, comprising the LamB secretion signal sequence fused to mature bovine somatotropin (bST), in Escherichia coli. Plasmid constructs with the recA promoter showed significant protein accumulation prior to induction and cell lysis occurred after induction. In contrast, the lacUV5 promoter was tightly regulated. With the lacUV5 promoter, temperature and inducer concentration had significant effects on the total amount of recombinant protein produced and the fraction processed to mature bST. Quantitation of bST from shake flask cultures showed that 1-2 micrograms/ml/OD550 could be released from the periplasm by osmotic shock. N-terminal sequence analysis of the purified protein indicated that the majority of the secreted bST was correctly processed. The bST present in the osmotic shock fraction was judged to be correctly folded by comigration with oxidized methionyl-bST standard on a non-reducing polyacrylamide gel and activity in a bovine liver radioreceptor assay. These results provide a rapid method to produce bST for use in structure-function studies.

Production of bovine insulin-like growth factor 2 (bIGF2) in Escherichia coli.

Bovine insulin-like growth factor 2 (bIGF2) was produced in inclusion bodies in the cytoplasm of Escherichia coli and accumulated at high levels: 20-25% of total Coomassie-stained bacterial protein. The level of accumulation of bIGF2 was affected by the choice of codons in the 5' end of the coding sequence and by a rpoH mutation in the host cells. Purified recombinant bIGF2 had the native N terminus and the same mitogenic activity as that of bIGF2 purified from bovine serum.

Expression of secreted insulin-like growth factor-1 in Escherichia coli.

The synthesis, processing and secretion of insulin-like growth factor-1 (IGF-1 or somatomedin-C) fused to LamB and OmpF secretion leader sequences in Escherichia coli have been investigated. Expression and secretion of IGF-1 was achieved. The major portion of this secreted IGF-1 accumulated in the periplasmic space as insoluble aggregates. A small amount of IGF-1 was found folded in its native conformation in the medium. The lamB and ompF signal sequences were fused to the 5' coding sequence of IGF-1. Fusion of the lamB signal sequence directly to IGF-1 (lamB-IGF-1) resulted in accumulation of 16-20 micrograms/A550/ml of correctly processed IGF-1 in the periplasmic space. The processing efficiency of LamB-IGF-1 and OmpF-IGF-1 was enhanced in an E. coli strain bearing a prlA4 mutation. Amino acid sequence analysis of IGF-1 secreted into the periplasm and exported into the medium confirmed the precise removal of the LamB or OmpF signal sequence. IGF-1 synthesized in E. coli was demonstrated to be active in a cell proliferation bioassay.

Mistranslation in IGF-1 during over-expression of the protein in Escherichia coli using a synthetic gene containing low frequency codons.

Partial misincorporation of Lys for Arg has been observed for the Arg residues of IGF-1 when the molecule is expressed in Escherichia coli using a synthetic gene with the low frequency AGA codon encoding all six Arg residues and yeast preferred codons encoding the remaining residues. The Lys for Arg substitution at these residues could not be detected when a gene containing E. coli preferred codons, with the codon CGT coding for all Arg residues, was used for the expression of the protein. Similarly, no misincorporation of Lys for Arg could be detected when a gene containing Escherichia coli preferred codons at all positions, except for an AGA codon at Arg (36), was utilized.

Translocation of the precursor of 5-enolpyruvylshikimate-3-phosphate synthase into chloroplasts of higher plants in vitro.

5-enolPyruvylshikimate-3-phosphate synthase (EPSP synthase; 3-phosphoshikimate 1-carboxyvinyl-transferase; EC 2.5.1.19) is a chloroplast-localized enzyme of the shikimate pathway in plants. This enzyme is the target for the nonselective herbicide glyphosate (N-phosphonomethylglycine). We have previously isolated a full-length cDNA clone of EPSP synthase from Petunia hybrida. DNA sequence analysis suggested that the enzyme is synthesized as a cytosolic precursor (pre-EPSP synthase) with an amino-terminal transit peptide. Based on the known amino terminus of the mature enzyme, and the 5' open reading frame of the cDNA, the transit peptide of pre-EPSP synthase would be maximally 72 amino acids long. To confirm this prediction and to assay directly for translocation of pre-EPSP synthase into chloroplasts in vitro, we cloned the full-length cDNA into an SP6 transcription system to produce large amounts of mRNA for in vitro translation. The translation products, when analyzed by NaDodSO(4)/PAGE autoradiography, indicate a relative molecular mass for pre-EPSP synthase of approximately 55 kDa. Uptake studies with intact chloroplasts, in vitro, indicate that pre-EPSP synthase was rapidly taken up into chloroplasts and proteolytically cleaved to the mature approximately 48-kDa enzyme. The transit peptide was shown to be essential for import of the precursor enzyme into the chloroplast. To our knowledge, post-translational import into chloroplasts of a precursor enzyme involved in amino acid biosynthesis has not been reported previously. Furthermore, enzymatic analysis of translation products indicates that pre-EPSP synthase is catalytically active and has a similar sensitivity to the herbicide glyphosate as the mature enzyme. To our knowledge, pre-EPSP synthase represents the only example of a catalytically competent chloroplast-precursor enzyme.

Electron microscopy of secondary structure in partially denatured precursor and mature Escherichia coli 16 S and 23 S rRNA.

The secondary structure of 16 S and 23 s rRNA sequences in 30 S preribosomal RNA of Escherichia coli was analyzed by electron microscopy after partial denaturation and compared to mature 16 S and 23 S rRNA examined under the same conditions. The sequences in the pre-rRNA notably lack the specific loops that dominate the 5'-terminal regions of mature 16 S and 23 S rRNA. In other respects, the sizes and locations of loops in the 23 S rRNA sequence are qualitatively very similar in mature and pre-rRNA. Eleven of 12 loops outside of the 5'-terminal domain correspond, with the most frequent features in the 3'-half of the molecule. In contrast, the sizes and locations of loops in the 16 S rRNA sequence differ between precursor and mature forms. In the pre-rRNA, instead of the 370-nucleotide 5'-terminal loop of mature rRNA, some 1000-nucleotide terminal loops are observed. The pre-rRNA also shows a frequent 610-nucleotide central loop and a large 1240-nucleotide loop not seen in the mature rRNA. Also, in the 3'-region of the sequence, the largest loops in pre-rRNA are 120 nucleotides shorter than in mature rRNA. We suggest that the structure of pre-rRNA may promote some alternate conformational features, and that these could be important during ribosome formation or function.

Alternative conformations in Escherichia coli 16S ribosomal RNA.

Partially denatured 16S rRNA from 30S ribosomes shows features of secondary structure in electron microscopy that correspond to the well accepted secondary structure model derived from chemical modification and phylogenetic data. However, a very different conformation is seen in precursor 16S rRNA sequences contained within 30S pre-rRNA transcripts: the major 5'-terminal loop is absent, and several additional quite stable large loops, symmetrically placed in the molecule, are present. Features of the alternative structure are also seen in mature 16S rRNA from Escherichia coli and from two Bacillus species when heated in certain buffers. Microscopy thus reveals specific features of alternative conformations and their relative stabilities, suggesting a possible transition during ribosome formation.

Electron microscopy of the secondary structure in partially denatured rRNAs of Escherichia coli and Bacillus stearothermophilus.

Partially denatured 16S and 23S rRNAs from the thermophile Bacillus stearothermophilus show characteristic loop patterns when observed by electron microscopy. The patterns are very similar to those seen in rRNAs from Escherichia coli. At least 2 of 4 most stable interactions in 16S rRNA and 8 of 12 interactions in 23S rRNA are in common for the two species. These interactions correspond well to features of secondary structure in models inferred for rRNA from phylogenetic sequence comparisons and chemical modification studies. However, two additional large loops, enclosing large portions of the 23S rRNA, have been detected in B. stearothermophilus for the first time, and even though other loops are similar, their relative frequencies vary in the two species. Much of the variation is consistent with relative delta G degree values for putative base-paired stems at the base of different loops; but the 5'-terminal loops in 23S rRNA, for example, are unaccountably far less stable in B. stearothermophilus. Also, in general, structural features are not differentially stabilized in B. stearothermophilus; the relative stability of secondary structure in its ribosomes at elevated growth temperatures must involve interactions with ribosomal proteins or other cellular components.

Electron microscopy of secondary structure in partially denatured Escherichia coli 16S rRNA and 30S subunits.

Loops observed in partially denatured 16S rRNA lie within three domains, each about 500 nucleotides long. The loops observed in the 5' and central domains agree well with features of the model proposed by Woese et al. [Woese, C. R., Gutell, R., Gupta, R., & Noller, H. F. (1983) Microbiol. Rev. 47, 621-669]. The structure in the 3' domain is more complex and variable but is still consistent with the model. Published psoralen cross-linking studies have reported one of the observed loops but have also identified loops other than those observed here or predicted by any secondary structure model. These loops are stabilized by increasing concentrations of Mg2+ ions and by bound ribosomal proteins. For example, protein S4 in LiCl core particles stabilizes a loop of 370 nucleotides which forms part of its putative binding site on rRNA. The loop structures are characteristic enough to permit an overall comparison of the most stable of the predicted and observed loops in 16S and 23S rRNAs. Both rRNAs show a stable 5'-terminal loop and a set of subterminal nested loops near the 3' end.

Structure of partially denatured Escherichia coli 23 S ribosomal RNA determined by electron microscopy.

The secondary structure of 23 S ribosomal RNA was analyzed by electron microscopy after partial denaturation. A reproducible pattern of loops was seen when molecules were spread for electron microscopy in 50% formamide solutions containing various concentrations of Mg2+ and Na+. Some loops were stabilized more than others by Na+ or by Mg2+; but in general, small amounts of Mg2+ (0.5 to 1.0 mM) markedly stabilized all the major loops, as did much greater amounts of Na+ (100 mM). However, at all levels of Mg2+ examined, increasing levels of Na+ destabilized loop structures. These data are consistent with the known salt dependence of double-stranded DNA and transfer RNA structure. The four most frequently observed loops correspond, within the limits of measurement error, to the major loops in the secondary structure models of Noller et al. (1981) and Glotz et al. (1981). These four loops are, in length and position of their midpoints along the 23 S rRNA molecule: 490 +/- 50 at 250 +/- 40; 350 +/- 50 at 1860 +/- 80; 400 +/- 70 at 2330 +/- 150; and 570 +/- 100 at 2350 +/- 100. Three of the four have base-paired stems with delta G0 values among the lowest of all the loops in the two indirect models. At least two are also among the most stable loops found in computer searches of the 23 S rRNA sequence for dyad symmetry. These results demonstrate that partial denaturation mapping can both identify prominent features of secondary structure in rRNA and estimate their relative stability.