Eosinophil-derived Neurotoxin and Human Liver Ribonuclease IDENTITY OF STRUCTURE AND LINKAGE OF NEUROTOXICITY TO NUCLEASE ACTIVITY*

Eosinophil-derived neurotoxin (EDN) and human liver RNase were found to be indistinguishable from each other but distinct from the pancreatic ribonucleases in their nucleolytic activity on polynucleotides or small defined substrates. Antibodies to EDN and liver RNase showed identical cross-reactivities in assays of nuclease inhibition and in a radioimmunoassay. In each instance, EDN and liver RNase were easily distinguished from bovine or human pancreatic RNase. When injected intrathecally into rabbits, 5-10 gg of EDN or liver RNase each was neurotoxic as judged by induction of the Gordon phenomenon. Human pancreatic RNase was less neurotoxic, and up to 20-fold higher levels of bovine pancreatic RNase showed no effect. of EDN, liver RNase, and eosinophil cationic protein with iodoacetic acid at pH 5.5 in inactivation of their RNase activity and also de-stroyed their neurotoxicity.

activity is necessary but not sufficient to induce the Gordon phenomenon.

EXPERIMENTAL PROCEDURES
Proteins-The purification of EDN (3), ECP (4), and human liver (16) and pancreatic (18) RNases and the production and characterization of antibodies to each protein have been described. Antibody preparations were partially purified by ammonium sulfate fractionation, followed by passage through a column of DEAE-cellulose and carboxymethyl cellulose in order to remove serum ribonucleases (22).
The site-specific modification of RNases with iodoacetic acid was adapted from the conditions used by Plapp (23): protein (0.18 mg/ ml) in 200 mM MES (adjusted to pH 5. 5 with NaOH), 167 mM NaCI, 15 mM sodium iodoacetate, was incubated at 37 ' in the dark for up to 24 h. In parallel incubations iodoacetate was omitted (buffer control) or 30 mg/ml reduced glutathione was added (sham carboxymethylation). Aliquots were removed periodically and assayed for residual RNase activity. Preparations of carboxymethylated protein, usually showing less than 5% residual nuclease activity, were then dialyzed against several changes of distilled water and lyophilized. Protein denaturation by reduction and carboxymethlyation of all cysteine residues has been described (24). Placental RNase inhibitor (RNasin) was obtained from Calbiochem and Promega, and Inhibit-ACE was purchased from 3' + 5' Inc.
Assays-Except when otherwise specified, RNase activity was quantitated through the formation of perchloric acid-soluble nucleotides from a wheat germ ribosomal RNA substrate under conditions optimized for the pancreatic (18) or liver (16) enzymes. One unit of activity in this procedure corresponds to about 2.5 ng of human pancreatic RNase or 8 ng of liver RNase. Other procedures were used for comparative purposes. The Kunitz (25) assay employed yeast RNA (Sigma) that had been further purified (26) to reduce amounts of small oligonucleotides. Spectrophotometric assays of the hydrolysis of polynucleotides, dinucleoside phosphates, and nucleoside cyclic 2',3'-phosphates have been described (16, 27-29); a temperaturecontrolled Beckman DU-50 spectrophotometer was used.
Assays of RNasin and Inhibit-ACE inhibitory activity were modified from a published method (26). The stability of RNasin-RNase complexes was measured using the approach of Lee et al. (30). Enzyme plus 1 or 10 eq of RNasin were incubated at 0 or 37 "C in a buffer containing 2 mg/ml bovine serum albumin plus 50 mM MOPS, pH 7.5. Samples were withdrawn at intervals of up to 50 h, and levels of uninhibited RNase activity were measured. As a control, RNase samples were incubated in the absence of RNasin under the same conditions and assayed at similar intervals.
Measurement of antibody inhibition of RNase activity involved incubation of varying amounts of antibodies with 1 unit of enzyme in enzyme assay buffer for 2 h on ice, followed by elevation of the temperature to 37 "C and RNase assay. Interference in this assay by reduced and carboxymethylated EDN or by EDN that had been inactivated at pH 5.5 with iodoacetic acid measured the restoration of antibody-inhibited RNase activity. One unit of RNase plus variable amounts of chemically modified EDN were mixed in assay buffer with sufficient antibody to inhibit 1 unit of RNase activity by 50%; samples were incubated for 15 min at 37 "C followed by 2 h on ice, and then assayed for RNase activity. The immunochemical reactivity of the RNases was also analyzed by a competitive binding doubleantibody radioimmunoassay described previously (31).
The biological assay for neurotoxic activity (the Gordon phenomenon) followed established methods (3,4). Rabbits that had been injected with 200 p l of protein in saline/citrate buffer (or buffer alone as a control) were observed daily for up to 12 days for forelimb stiffness, incoordination, and ataxia. Animals were evaluated by an individual who was not aware of the specific sample injected in each case and were scored positively if all of the symptoms described above were apparent.

RESULTS
Actioity as RNases-EDN and liver RNase were compared with each other and with bovine pancreatic RNase A in assay procedures that used both macromolecular and low molecular weight substrates. As shown in Table I, with RNA and polynucleotide substrates EDN and liver RNase were essentially identical and clearly unlike RNase A. Under the conditions used EDN and liver RNase both showed a preference for polynucleotides that contained uridine rather than cytidine, while the opposite was true for pancreatic RNase. Both liver RNase and EDN showed an identical and generally low activity with simple polymeric substrates, whether single-stranded (poly(U) or poly(C)) or double-stranded (poly(A) .poly(U) or poly(I).poly(C)). With small defined substrates EDN and liver RNase were again essentially identical; none of these compounds was hydrolyzed efficiently. Again this is in contrast to RNase A, which was reasonably effective in the hydrolysis of several small substrates.
Immunological Cross-reactivity"Polyclona1 antibodies to human pancreatic RNase (18), t o E D N (3), and to liver RNase (16) were previously characterized in our laboratories. These antibody preparations were examined for their ability to inhibit the nucleolytic activities of E D N a n d liver RNase. Antibodies to either EDN or liver RNase inhibited the RNase activity of both proteins to an equivalent extent, as shown in Fig. 1. As expected from earlier results (16) inhibition of human pancreatic RNase by antibodies to EDN or liver RNase was not detectable at the antibody levels used in these experiments. Similarly, in radioimmunoassays using lz5I-labeled liver RNase as the radioligand, binding by either anti-E D N or anti-liver RNase was inhibited to an equivalent extent by an equal quantity of either EDN or liver RNase (results not shown). Antibodies to human pancreatic RNase show less that 1% cross-reactivity with human liver ribonuclease in a similar assay (16).
Inactivation and Inhibition of RNase Activity-When samples of E D N a n d liver RNase were incubated at pH 5.5 with iodoacetic acid, each protein lost nuclease activity at an identical rate (Fig. 2) that was essentially the same as the rate of inactivation of bovine and human pancreatic RNases (16, 18). In order to test whether protein conformation is affected by carboxymethylation, we investigated the crossreactivity of iodoacetate-modified protein with polyclonal antibodies t o E D N a n d liver RNase. As shown in Fig. 3, carboxymethyl-EDN was still an effective competitor in a radioimmunological assay for EDN using antibodies raised against either protein. In contrast, fully reduced and carboxymethylated EDN possessed less than 0.01% of its initial reactivity.
Carboxymethyl-EDN (but not EDN that was denatured by reduction and carboxymethylation of its cysteines) also effectively competed with native EDN in assays based on measurement of antibody inhibition of RNase activity. As shown in Fig. 4 nuclease inhibition by anti-EDN or anti-liver RNase antibodies was relieved by iodoacetate-inactivated EDN, while reduced and carboxymethylated protein had no effect on RNase activity measured in these experiments.
RNasin, a placental protein that efficiently inhibits pancreatic RNase activity, also inhibits EDN and liver RNase ( Fig. 5 ) . A different inhibitory protein, Inhibit-ACE, showed no effect on any of the enzymes of human origin, but less than 0.1% as much inhibitor strongly inhibited bovine pancreatic RNase. EDN and liver RNase were thus indistinguishable from one another in these experiments.
Neurotoxic Actiuity: Induction of the Gordon Phenomenon-The results of experiments comparing EDN and liver RNase as neurotoxins and of attempts to dissociate their nucleolytic and neurotoxic activities are summarized in Table 11. Experiment 1 directly compared EDN and liver RNase as neurotoxins; both were active in induction of the Gordon phenomenon, and the two proteins functioned at comparable doses. The latter conclusion is also based on other observations included elsewhere in Table 11. Experiment 2 demonstrates the linkage of neurotoxic and nuclease activities; RNase activity was reduced t o 2-5% of normal by treatment with iodoacetic acid at p H 5.5; carboxymethyl (CM)-EDN, liver CM-RNase, and CM-ECP were all devoid of neurotoxic activity. Controls, all of which were neurotoxic, included native proteins, EDN that had been exposed to the buffer and incubation conditions of the experiment but not iodoacetate (buffer control), EDN  treated with iodoacetic acid in the presence of excess 2mercaptoethanol (sham CM-EDN), and EDN that was simply shipped between laboratories with the other samples. In contrast, EDN that had been denatured by reduction and carboxymethylation was inactive in this assay. Experiment 3 of Table I1 shows that RNase activity alone is not sufficient to confer neurotoxicity. Neither bovine RNase A nor its glycosylated equivalent RNase B induced the Gordon phenomenon, even at levels 10-20-fold higher than were effective with EDN or liver RNase. Human pancreatic RNase gave a positive response, but only at a dosage that was significantly higher than that found effective with EDN or liver RNase. It is known that the human pancreatic RNase preparation contains both glycosylated and nonglycosylated forms of the enzyme (321, and we speculated that only one form might be neurotoxic. This may be true; although both forms showed RNase activity, the nonglycosylated (unadsorbed to concanavalin A-Sepharose) rather than the glycosylated ( i e . acideluted) fraction was active in the assay. However, the acideluted fraction readily precipitated from solution and lost  Table I1 we attempted to neutralize EDN with a small excess of the nuclease inhibitor protein RNasin, and found no inhibition.
However, we were concerned that the inhibitor or the EDN-RNasin complex might be unstable and release RNase during the assay. We tested this hypothesis by directly measuring the release of RNase activity from complexes that were formed using either sufficient RNasin to inhibit about 90% of the nuclease activity or a 10-fold excess of inhibitor. At 0 "C the apparent release of RNase activity during the 50 h of incubation was no more than 15%, regardless of the RNasin level. At 37 "C 10-15% of the RNase was detectable after 30 and 50 h of incubation with the 10-fold excess of RNasin. More important to Experiment 4 of Table 11, incubation at 37 "C of EDN with a nearequivalent level of RNasin was accompanied by the release of about 20% of the RNase activity after 3 h and almost total 2 .@ the RNase activity (of control, uncompeted samples) by 50% was added, and samples were incubated for 15 min at 37 ' followed by at least 2 h on ice. The reaction was initiated by the addition of the wheat germ RNA substrate. The data are presented as the release of RNase from antibody inhibition. Open symbols, anti-EDN; closed symbols, antiliver RNase; A, native EDN as competitor; 0, CM-EDN as competitor; 0, reduced and carboxymethylated EDN as competitor.

DISCUSSION
Identity of EDN and Liver RNase-Our results strengthen the supposition that EDN and liver RNase are the same entity. It was already known that the NH2-terminal amino acid sequences of liver RNase (16) and EDN (4) are identical to each other and to the equivalent section of the full sequence of the nonsecretory RNase isolated from urine (17), and that the EDN cDNA sequence (19, 20) corresponds to the amino acid sequence of the urinary RNase. Our inability to differentiate EDN and liver RNase in immunological measurements suggests that their entire sequences are probably identical and, in addition, that they are not likely to differ significantly in post-translational processing, including glycosylation; differences in these characteristics would be expected to affect antibody recognition, yet the proteins are indistinguishable by both RIA and activity inhibition measurements with two different polyclonal antibody preparations. Their sequence clearly places EDN and liver RNase within the superfamily of RNase proteins that also includes the pancreatic RNases (32, 33), ECP (34,35), and a protein that induces the formation of blood vessels in solid tumors, angiogenin (36). Like bovine or human pancreatic RNase, EDN and liver RNase cleave polyribonucleotides that contain pyrimidine residues but purine homopolymers are resistant to hydrolysis. Both EDN and liver RNase prefer uridine-containing polymers, and both show low activity (compared with pancreatic RNase) with dinucleoside phosphate substrates. Their action on polynucleotides indicates that secondary structure may be particularly important in substrate selection since only RNA and single-stranded homopolymers are easily hydrolyzed. These catalytic properties place EDN and liver RNase in the "cellular" or nonsecretory group of nucleases, rather than in the secretory group of enzymes typified by bovine pancreatic RNase (33,37). ECP has also been placed in the nonsecretory category on the basis of its enzymatic characteristics (29). Angiogenin has very limited RNase activity (38), but its amino acid sequence is more similar to turtle or human pancreatic RNases than to the human nonsecretory enzyme (17).
Imposition of the EDN/liver RNase sequence on the known structure of bovine RNase A as shown in Fig. 6 (see also Refs. 17, 33, and 39) shows conservation of the major residues at the active site of the enzyme and a clustering of positively charged residues at the protein surface but also somewhat distant from the cleft. The charged amino acids could contribute to polynucleotide binding, while the relative absence of protonated residues nearer the active site may explain decreased affinity for smaller substrates and help account for an inability to melt and attack double-stranded polynucleotides.
Additional evidence that the active site of liver RNase/ EDN is similar to that of pancreatic RNase comes from enzyme inactivation and inhibition. Reaction of bovine RNase A with iodoacetic acid at pH 5.5 results in carboxymethylation at either histidine residue 12 or 119 of the catalytic site of the enzyme (23, 40), and there is extensive evidence that the overall conformation of the protein is conserved (41). The similar kinetics of inactivation of EDN or liver RNase by iodoacetate (Fig. 2), the apparent retention of secondary  Positive responses/ total rabbits Glutathione present during iodoacetate treatment.
* Protein shipped between laboratories but not included in experimental manipulations.
Iodoacetate omitted from reaction mixture but sample carried through the experimental procedure. structure by the carboxymethylated proteins (Figs. 3 and 4), and the appropriate incorporation of radiolabeled iodoacetate (16) indicate that the inactivation involves a similar reaction that results in the modification of EDN/liver RNase histidine residues 15 and 129. Inhibition of the nuclease activity of EDN/liver RNase by RNasin is similar to that seen with human pancreatic RNase (Fig. 4) (26, 44). Calculation of the quantities of enzyme and inhibitor used to generate the data of Fig. 4 indicates that EDN and liver RNase are also inhibited in the form of equimolar complexes with RNasin. Measurements of complex stability suggest that the affinity of liver RNase/EDN for RNasin is similar to that found with pancreatic RNase A (26) or human pancreatic RNase, but possibly much lower than that reported for angiogenin (30) or a ribonuclease from placenta that appears closely related to the nonsecretory enzymes (43).

Neurotoxic Activity and Its Dependence on RNase Activity-
The data shown in Table I1 show that liver RNase is capable of inducing the Gordon phenomenon and that the potency of the liver enzyme and EDN are comparable. This is additional evidence that these are identical proteins. The data also link neurotoxic action to RNase activity; human pancreatic RNase is effective in this assay, although smaller amounts of EDN/ liver RNase or ECP are required relative to the pancreatic enzyme. However, RNase activity is not sufficient to induce the Gordon effect; even much larger quantities of bovine pancreatic RNase A or B had no effect.
Two characteristics in which these proteins differ (glycosylation and surface charge) could affect their ability to enter brain cells and exert nuclease activity. The nonglycosylated form of human pancreatic RNase was toxic, while the concanavalin A-reactive glycosylated fraction was not (Table 11). However, this observation is complicated by the observation that the glycosylated fraction of the enzyme appeared to aggregate and lose RNase activity; since the assay of the Gordon phenomenon takes several days, activity could have been lost in the process. It is also significant that liver RNase (16) , EDN (4), and ECP (4, 45) are all neurotoxic glycoproteins, while both the glycosylated and nonglycosylated forms of bovine pancreatic RNase did not induce the Gordon phenomenon. It is thus difficult to directly link glycosylation and neurotoxicity. The basicity of these proteins is compared in FIG. 6. Amino acid sequence of EDN/liver RNase superimposed on the bovine RNase structure. e, active site residues.
Unnumbered circles represent additional residues occurring before bovine RNase amino acid 1 and after residues 69, 91, and 113. The broken line between residues 16 and 23 represents a deleted sequence. Charged residues are indicated by a + (Lys, Arg), -(Glu, Asp), or & (His); each terminus is also marked with a charge. A, sites of glycosylation. Table 111. Bovine RNase A has the lowest net positive charge and is not neurotoxic. Human pancreatic RNase is slightly more basic and is somewhat toxic; glycosylation would decrease this charge if sialic acid residues were present, possibly explaining our results with human pancreatic RNase fractions. Liver RNase/EDN are slightly more basic and are neurotoxic a t lower levels. ECP is much more basic; it is less active as an RNase (4, 29), but at least as neurotoxic as EDN in rabbits (4) and more toxic in guinea pigs (45). Angiogenin, intermediate in basicity between EDN and ECP, is a poor nuclease (38); it would be interesting to know if angiogenin is able to induce the Gordon phenomenon.
The most exciting observation in Table I1 is that controlled inactivation of the RNase activity of EDN with iodoacetic acid also inactivates its neurotoxic action. The conformation of bovine RNase A is not greatly altered in this reaction (41). The ability of carboxymethylated EDN to efficiently react with antibodies and compete in both radioimmunological (Fig.  3) and RNase inhibition assays (Fig. 4), while the unfolded reduced and carboxymethylated EDN does not compete, indicates that the conformation of carboxymethyl EDN is also retained. The assays of Figs. 3 and 4 also measure somewhat different aspects of EDN conformation. The radioimmunological assay involves antigenic determinants at many sites on the protein surface, while antibody inhibition of RNase activity primarily involves regions that are involved in substrate binding. Quantitative evaluation of the radioimmunological assay of Fig. 3 or the RNase inhibition assays of Fig.  4 indicates that more than 50% of the CM-EDN is fully crossreactive with each preparation of antibodies, yet Table I1 shows that 5 mg of native EDN is neurotoxic while 4 times as much CM-EDN is not. We thus conclude that there is a direct linkage between RNase activity and neurotoxic function.
Angiogenin, a protein of the RNase superfamily that induces vascularization of solid tumors, is also a nuclease with very limited activity (36). Both activities are lost upon controlled carboxymethylation with bromoacetic acid (36,46,47). Moreover, both activities are enhanced by modification of angiogenin at aspartate residue 116 (48). These results link angiogenesis closely to RNase activity. Other data (49) indicate that a peptide segment (residues 61-67) that is not involved in nuclease action is also required for angiogenesis. This peptide segment is part of the base of the molecule if it is oriented as EDN is in Fig. 6, and it is in a region that includes several charged residues (see also Ref. 16). It may be significant that the region Of EDN (residues num-13. Kaufman, L. D., Gruher, B., and Gregersen, P. K. tribute the absence of effect of RNasin in Table 11 to