Translational autocontrol of the Escherichia coli hfq RNA chaperone gene

INTRODUCTION

The Hfq protein was first recognized in Escherichia coli as a host factor required for replication of phage Qβ (Franze de Fernandez et al. 1968). However, the importance of Hfq in cellular physiology was acknowledged just a decade ago when the broadly pleiotropic phenotypes of an E. coli hfq mutant were characterized (Tsui et al. 1994). The observations that Hfq is involved in expression of the rpoS gene encoding the stationary sigma factor, σ³⁸ (Brown and Elliot 1996; Muffler et al. 1996), stability control of several mRNAs (Tsui et al. 1997; Vytvytska et al. 2000; Folichon et al. 2003) and small regulatory RNAs (sRNAs) (Sledjeski et al. 2001; Moll et al. 2003a), post-transcriptional regulation of target mRNAs by sRNAs (Sledjeski et al. 2001; Massè and Gottesman 2002; Møller et al. 2002; Zhang et al. 2002; Večerek et al. 2003), that it acts as an RNA chaperone on mRNAs (Moll et al. 2003b; Geismann and Touati 2004), and as a virulence factor in several bacterial pathogens (Robertson and Roop 1999; Sonnleitner et al. 2003; Christiansen et al. 2004; Ding et al. 2004) have recently sparked a greater than ever interest in this highly conserved bacterial protein (for review, see Valentin-Hansen et al. 2004).

Electron microscope studies of the E. coli Hfq protein (Møller et al. 2002; Zhang et al. 2002) and X-ray crystallography of the Staphylococcus aureus (Schumacher et al. 2002) and Pseudomonas aeruginosa (Nikulin et al. 2005) Hfq homologs, as well as of the amino-terminal 72 amino acids of E. coli Hfq (Sauter et al. 2003) showed that it has a hexameric ring-shaped structure, and that it belongs to the large family of Sm-like proteins. These proteins primarily recognize short U-rich stretches (Achsel et al. 2001) and are involved in RNA processing in eukaryotic cells. An A/U-rich region preceded or followed by a stem–loop structure has similarly emerged as a common binding motif for Hfq (Brescia et al. 2003; Moll et al. 2003a; Geissmann and Touati 2004).

The Hfq protein forms hexamers in solution (Arluison et al. 2002) and is highly abundant (~10,000 Hfq-hexamers per cell; Valentin-Hansen et al. 2004), which in turn seems to account for its ability to affect multiple targets in the cell. In contrast to the Hfq synthesis rate, which was reported to be enhanced in exponentially growing cells when compared to stationary phase (Kajitani et al. 1994), it has also been reported that the total concentration of Hfq is higher in slow growing cells (Vytvytska et al. 1998) and upon entry into stationary phase (Tsui et al. 1997). The higher abundance of Hfq in stationary phase parallels the increased expression of sRNAs (Argaman et al. 2001; Zhang et al. 2003), many of which are targeted by Hfq. The E. coli hfq gene is part of the amiB-mutL-miaA-hfq-hflX-hflK-hflC operon, whose transcription is complex and driven by several promoters (Tsui and Winkler 1994). An internal heat-shock promoter serves the hfq gene, which appears to ensure the maintenance of high level hfq transcription under stress conditions (Tsui et al. 1996).

Recent work revealed that Hfq binds to the 5′-end of ompA mRNA, encoding the outer membrane protein A, and changes its structure in a manner which is detrimental to translation initiation complex formation (Vytvytska et al. 2000; Moll et al. 2003b). Similarly, Hfq was shown to inhibit translation of sodB mRNA by mediating its interaction with the sRNA RyhB (Geissmann and Touati 2004). In both cases, the block of translation results in functional inactivation of the mRNA in an RNase E-dependent manner (Vytvytska et al. 2000; Massé et al. 2003). Given the important roles of Hfq in RNA metabolism, in this study we have addressed the question whether Hfq synthesis is subject to an autoregulatory circuit, wherein Hfq controls its own synthesis at the translational level.

RESULTS AND DISCUSSION

Hfq binds to two sites in the 5′-UTR of its mRNA

Hfq has been reported to destabilize its own mRNA in an RNase E-dependent manner, suggesting that hfq expression is autogenously regulated at the post-transcriptional level (Tsui et al. 1997). Therefore, we first employed gel-mobility shift assays to test whether Hfq binds to the 5′-UTR of its own transcript. As depicted in Figure 1A, hfq126 RNA comprises the entire 5′-UTR up to nucleotide +57, including two computer- predicted (Mathews et al. 1999) stem–loop structures with ΔG values of −5.8 (h1) and −15.6 kcal/mol (h2). Hfq binding to hfq126 RNA resulted in two shifted bands, which suggested the presence of two distinct Hfq binding sites (Fig. 1B, panel I).

Hydroxyl radical footprinting was then used to map the Hfq-binding sites in the 5′-UTR of hfq126 RNA. Two sites were protected by Hfq from hydroxyl radicals, the U-rich stretch at position −55 to −50 (Fig. 1A, C; Hfq binding site A) as well as the region between nucleotides −20 to +4 (Fig. 1A, C; Hfq binding site B), encompassing the ribosome binding site (RBS).

It has been reported that efficient Hfq binding to RNA requires stem–loop structures in the vicinity of the primary binding sequence (Brescia et al. 2003; Moll et al. 2003a; Geissmann and Touati 2004). Therefore, we tested by enzymatic probing with ribonuclease T1 whether both stem–loop structures shown in Figure 1A are present in hfq mRNA (nucleotides −68 to +102). As shown in Figure 1D, T1 cleavage of this mRNA followed by subsequent primer extension revealed that the G residues in h1 and h2 were completely protected, indicating that both stem–loop structures are stable elements of the 5′-UTR of hfq mRNA.

Next, band shift assays were performed with hfq126 RNA derivatives (Fig. 1A) lacking either the 5′-terminal part (hfq96 RNA) or the immediate coding region including h2 (hfq76 RNA). Binding of Hfq to hfq96 RNA resulted in only one shifted band in the gel-mobility shift assay (Fig. 1B, panel II) consistent with the absence of Hfq binding site A on this RNA. Binding of Hfq to hfq76 RNA gave rise to mainly one shifted band as well (Fig. 1B, panel III), whereas the band likely resulting from two bound Hfq-hexamers was hardly discernible. The hfq76 RNA is lacking h2, which could affect Hfq binding to site B. To test this further, band shifts were also performed with hfq46 RNA (Fig. 1A), which is devoid of the up stream Hfq binding site A as well as of h2. As shown in Figure 1B, panel IV, Hfq did not bind to this RNA indicating that Hfq binding to site B requires the stem–loop structure. Since we did not observe a significant difference in the binding affinity of Hfq for either site on the hfq96 or the hfq76 transcript (Fig. 1B, panel II, III), these gel-mobility shift assays suggested that both sites are occupied by Hfq without preference, rather than in a cooperative manner. Moreover, it seems likely that the faster migrating hfq126 complex (Fig. 1B, panel I) is a mixture of hfq126 transcripts, where Hfq is bound either to site A or B.

Autogenous repression of hfq mRNA translation

Translational repressors inhibit formation of the translation initiation complex between 30S ribosomes, initiator-tRNA, and mRNA. This can occur by direct competition with 30S subunit binding (Romby and Springer 2003), as in the cases of R17 coat protein, T4 gene 32 protein, T4 regA protein, and threonyl-tRNA synthetase, or by entrapment of the 30S subunit in an unproductive binary complex with mRNA, as exemplified by ribosomal proteins S15 and S4 (Philippe et al. 1993; Schlax et al. 2001).

To test whether Hfq acts as an auto-genous translational repressor, we first used an E. coli S30 in vitro translation system and programmed it with both full length hfq and ppiB (control) mRNAs in the presence of increasing concentrations of Hfq. The ppi mRNA, encoding rotamase B, was used as a control because we had recently shown that Hfq neither affects in vitro translation initiation complex formation nor in vitro translation of this mRNA (Večerek et al. 2003). The translation of hfq mRNA decreased concomitantly with increasing amounts of Hfq and was significantly repressed when Hfq was added in an eightfold (Hfq-hexamer) molar excess over both mRNAs (Fig. 2A). In contrast, translation of ppiB mRNA continued unabated (Fig. 2A), indicating that Hfq acts specifically on its own mRNA.

Next, we used an in vitro toeprinting assay (Hartz et al. 1988) to test directly whether Hfq prevents translation initiation complex formation on its own mRNA by impeding ribosome binding. Hfq was added to a hfq transcript spanning nucleotides −68 to +102 prior to the addition of ribosomal subunits and initator-tRNA at different molar ratios of Hfq (Hfq-hexamer) to mRNA. As shown in Figure 2B, the toeprint signal obtained after extension of primer Y19 decreased with increasing concentrations of Hfq, and formation of the 30S translation initiation complex was strongly diminished when Hfq was added in a 20-fold molar excess to mRNA. Previous toeprinting experiments conducted with E. coli lpp mRNA (Vytvytska et al. 2000) and, as mentioned above, with ppiB mRNA (Večerek et al. 2003) demonstrated that Hfq did not inhibit ribosome binding per se—i.e., Hfq has no general detrimental effect on translation initiation. We therefore considered this inhibitory effect of Hfq to be specific for hfq mRNA rather than simply reflecting the high affinity of Hfq for RNA (Senear and Steitz 1976).

Since the band shift experiments (Fig. 1B) implicated the stem–loop structure h2 (Fig. 1A) in Hfq binding to site B, which includes the ribosome binding site, we also asked whether h2 is of importance for translational auto-repression of hfq mRNA. We tested this possibility by performing a toeprinting experiment on hfq mRNA after annealing of oligonucleotide F22. F22 is complementary to the 3′ part of the stem of h2 (see Fig. 1A). Hence, binding of this oligonucleotide to hfq mRNA should result in melting of the structure. The occurrence of an extension signal—i.e., the generation of a cDNA by reverse transcriptase primed by oligonucleotide F22 (Fig. 2C, lane 1)—demonstrated that the oligonucleotide was bound to hfq mRNA (nucleotides −68 to +102). Consistent with the idea that Hfq binding to site B depends on h2, the protein did not inhibit ribosome binding on hfq mRNA annealed to oligonucleotide F22. Even a 20-fold molar excess of Hfq (hexamer) to mRNA resulted in a strong toeprint signal primed with oligonucleotide F22 (Fig. 2C, lane 3). In other words, in the absence of the stem– loop structure Hfq was apparently unable to abolish translation initiation complex formation.

To verify these in vitro data, the inducible plasmid-born hfq131-lacZ reporter gene (see Fig. 1A) was constructed, wherein the first 131 nt of hfq mRNA were fused to the eighth codon of the lacZ gene. The synthesis of the HfqΦLacZ protein was monitored in a hfq⁺ (MC4100F′) and in a hfq⁻ (AM111F′) genetic background. The β-galactosidase activities were determined in each strain from samples taken 30 min after induction of the hfq131-lacZ gene. To account for possible variables, such as mRNA levels and/or mRNA stability or plasmid copy numbers in the hfq⁺ and hfq⁻ strains, the β-galactosidase activities were normalized to the respective hfq131-lacZ mRNA concentration, which in turn was normalized to 5S ribosomal RNA. As shown in Figure 2D, in the hfq⁺ strain, the relative translational efficiency of the hfq131-lacZ mRNA was only −44% of the synthesis obtained in the hfq⁻ strain. Taking into consideration this in vivo experiment and the in vitro experiments (Fig. 2A, B), we concluded that Hfq acts as an autogenous translational repressor.

Both Hfq binding sites A and B contribute to translational auto-repression

Next, we asked whether both binding sites in the 5′-UTR of hfq could act synergistically in translational auto-repression. We first compared the translational yield of equimolar concentrations of hfq wild-type mRNA and hfqΔ30 mRNA (Fig. 1A), which lacks Hfq binding site A, in the presence of increasing amounts of Hfq in an in vitro translation assay. As shown in Figure 3A, Hfq repressed translation of both mRNAs to a different extent. An ~1.5-fold and a threefold molar excess of Hfq (hexamer) over hfq wild-type mRNA and hfqΔ30 mRNA, respectively, was required to reduce the translational yield to 50%. Hence, hfqΔ30 mRNA was ~twofold less sensitive to Hfq repression when compared with wild-type mRNA, implicating Hfq binding site A and h1 in translational auto-repression. Hydroxyl radical footprinting of hfq96 mRNA was then performed in the presence of Hfq to demonstrate that the deletion of site A did not change the structure of the hfq 5′-UTR and that Hfq still binds to site B. As shown in Figure 3B, this experiment did not reveal any changes in the protection of binding site B by Hfq (see Fig. 1C). We therefore interpreted the in vitro translation data as showing that both sites are required for efficient translational repression of hfq mRNA.

For verification, we compared the relative translation rate of hfq131-lacZ mRNA (Fig. 1A) with that of hfq101-lacZ mRNA (Fig. 1A) in vivo. The hfq101-lacZ mRNA corresponds at the 5′-end to hfqΔ30 mRNA, and thus lacks Hfq binding site A. As mentioned above, to account for variables in both genetic backgrounds the β-galactosidase activities obtained with the hfq101-lacZ construct in a hfq⁺ and a hfq⁻ strain, respectively, were again normalized to the respective mRNA levels, which in turn were normalized to 5S rRNA. There was no significant difference in the relative translational efficiency of hfq131-lacZ mRNA and hfq101-lacZ mRNA in the hfq⁻ strain (data not shown). However, as shown in Figure 3C, the expression of the hfq101-lacZ gene was significantly less affected than that of the hfq131-lacZ gene (see Fig. 2D) in the hfq⁺ strain when compared to the hfq⁻ strain.

Taken together, the results shown in Figure 3 implicated both Hfq binding sites A and B in translational repression, although only binding site B overlaps with the RBS of the hfq mRNA. In addition, the stem–loop structure h2 in the immediate coding region is apparently required for Hfq binding to site B (Figs. 1B, 2C). How can we account for the contribution of Hfq binding site A in translational autorepression? Biochemical studies have suggested that two Hfqhexamers can form a dodecamer (Arluison et al. 2002), possibly through interactions of the non-polar surface of each hexamer (Schumacher et al. 2002). The Hfq-dodecamer would have RNA binding site(s) on either surface, and thereby could interact with the two Hfq binding sites A and B. In a simplistic view this Hfq-dodecamer complex could be more stable and thus inhibit ribosome loading more efficiently than a hexamer bound to site B alone. Similarly to Hfq, the E. coli global regulator CsrA has been demonstrated to interfere with translation initiation on glgC (Baker et al. 2002) and cstA (Dubey et al. 2003) by binding to two and three or four binding sites, respectively. Although the contribution of the respective CsrA binding sites to translational repression remains to be elucidated, in either case CsrA–CsrA interactions have been suggested to contribute to complex formation with the respective mRNAs.

The steady state level of hfq mRNA was shown to depend on the major E. coli RNA endonuclease, RNase E (Tsui and Winkler 1994). Although the levels of RNase E were reportedly unchanged in a hfq⁻ strain, the half-life of hfq mRNA was 5.8 min more than threefold higher in a hfq⁻ mutant than in the isogenic hfq⁺ strain (1.8 min; Tsui et al. 1997). The results obtained in this study strongly suggest that Hfq inhibits ribosome binding on its mRNA, and thus subsequent transit of ribosomes through the coding sequence, which should render the untranslated mRNA vulnerable to RNase E cleavage and would be expected to result in rapid functional inactivation of the mRNA (Iost and Dreyfus 1995; Baker and Mackie 2003). We have mapped in vitro several RNase E cleavage sites in the immediate coding region of hfq mRNA (I. Moll, unpubl.), which would rationalize such a pathway for hfq mRNA decay upon translational repression. Moreover, we have noticed that ribosome loading on hfq mRNA in the presence of Hfq is only inhibited when Hfq is added prior to 30S subunits (I. Moll, unpubl.). Thus, like for ompA mRNA (Vytvytska et al. 2000) mutually exclusive binding of either Hfq or ribosomes to hfq mRNA appears to determine the fate of hfq mRNA.

MATERIALS AND METHODS

Bacterial strains and growth conditions

The Escherichia coli strains MC4100F′ (Steiner et al. 1993) and AM111F′ (MC4100F′ hfq1:: Ω; 33) have been described. They were grown in Luria-Bertani medium (Miller 1972) supplemented with ampicillin (100 μg/ml) where appropriate to maintain selection of plasmids.

Construction of plasmids

The plasmids pUhfqwt and pUhfqΔ30 used as templates for in vitro mRNA synthesis were constructed as follows. The hfq gene was placed under transcriptional control of the T7 Φ 10 promoter by means of PCR. The forward primers A24 (5′-GCTCTAGATAATACGACTCACTATAGGGTATCGTGCGCAATTTTTTCAGAATCGAAAGGTTC-3′) and K25 (5′-GCTCTAGATAATACGACTCACTATAGGTTCAAAGTACAAATAAGCGTGTGAGGAAAAGAGAG AATG-3′) contained a XbaI site (bold) and the T7 promoter sequence (underlined) either followed by the 5′-terminal part of the 5′-untranslated region (UTR) of the hfq gene (corresponding to the transcriptional start of hfq mRNA when transcribed from the P3 promoter; Tsui et al. 1996) (A24) or starting 30 nt downstream of the 5′-end (K25). The sequence of the hfq reverse primer B19 (5′-GGAATTCCCGTGTAAAAAAACAGCCCGAAAC-3′) containing an EcoRI site (bold) is complementary to the sequence following the stop codon of the hfq gene. The PCR products obtained with primers A24/B19 and K25/B19 were cleaved with XbaI and EcoRI, and ligated into the corresponding sites of plasmid pUC18, resulting in plasmids pUhfqwt and pUhfqΔ30, respectively.

Plasmids pRhfq131 and pRhfq101, which bear inducible hfqlacZ translational fusion genes are derivatives of plasmid pRB381 (Brückner 1992). The PCR fragments comprising the lac promoter (from nucleotides −60 to +32) from plasmid pUHE21-2 (Lanzer and Bujard 1998) and the first 131 nt (−68 to +63) or 101 nt (−38 to +63) of the hfq gene were inserted into the SalI and BamHI sites of plasmid pRB381. In the resulting plasmids, the corresponding hfq-lacZ mRNAs, which contain the first 21 codons of the hfq gene fused to the eighth codon of the lacZ gene, are transcribed from the lac promoter.

β-galactosidase assays

Strains AM111F′ (hfq⁻) and MC4100F′ (hfq⁺) harboring plasmids pRhfq131 (hfq131-lacZ) or pRhfq101 (hfq101-lacZ) were incubated at 37°C. At an OD₆₀₀ of 0.4, the plasmid encoded hfq-lacZ genes were induced by addition of IPTG (2 mM). Samples were with-drawn 30 min after induction to measure the β-galactosidase activities and for determination of the respective hfq-lacZ mRNA levels. The β-galactosidase activity was determined from triplicate samples as described (Miller 1972). The respective hfq-lacZ mRNA concentrations were determined by spotting total mRNA isolated from samples on a nitrocellulose-membrane followed by hybridization with a 5′-end [³²-P]-labeled lacZ-specific probe (5′-TTTCCCGG GATCCCGTCGTTTTACAACGTCGTGACTGGGAA-3′). The concentration of 5S rRNA was likewise determined using a 5′-end [³²-P]-labeled 5S RNA probe (5′-GGTGGGACCACCGCGCTACGGCCGC CAGGC-3′) and served as an internal control. The signals were visualized by a PhosphorImager (Molecular Dynamics) and quantified by ImageQuant software. The relative β-galactosidase values shown in Figures 2 and 3 were obtained by normalization of the different β-galactosidase values to the amount of the respective hfqlacZ mRNAs. Two independent sets of experiments were performed.

RNA constructs used in in vitro studies

For hfqmRNA synthesis, the plasmids pUhfqwt and pUhfqΔ30 were used as templates for in vitro transcription with T7 RNA polymerase (Promega). To prepare full length hfq mRNA, hfq126 mRNA, and hfq76 mRNA, the pUhfqwt plasmid was cleaved with EcoRI, AflIII, and DdeI, respectively. Cleavage of plasmid pUhfqΔ30 with the same enzymes yielded templates for synthesis of hfqΔ30, hfq96, and hfq46 mRNAs. The run-off transcripts were purified on 6% polyacrylamide-8M urea gels following standard procedures. The mRNA concentration was determined by measuring the A₂₆₀.

Gel shift assays

Gel-purified mRNAs were 5′-end labeled with [γ-³²P]-ATP (Amersham Pharmacia Biotech) and again purified on 6% polyacrylamide- 8M urea gels. Labeled mRNAs (5 nM) were incubated with or without increasing amounts of purified Hfq protein (as indicated in the legend to Fig. 2B) in a 10 μL reaction in binding buffer (10mM Tris at pH 7.5, 60mMNH₄Cl, 5 mM β-mercaptoethanol, 2mMMgOAc, 100 ng of yeast tRNA) for 5min at 37°C and then for 10 min at 0°C. The samples were then mixed with 40% glycerol to a final concentration of 10% and loaded on a native 4% polyacrylamide gel. Electrophoresis was performed in TAE buffer at 60 V for 12 h. Radioactive bands were visualized using a PhosphorImager.

Toeprinting analysis

The hfq-specific mRNA used for toeprinting was obtained as follows. First, a PCR with primers A24 (5′-GCTCTAGATAATAC GACTCACTATAGGGTATCGTGCGCAATTTTTTCAGAATCGA AA-3′) comprising the phage T7 Φ 10 promoter and Y19 (5′-CC CTTGCAGCTT-3′; complementary to nucletides +91 to +102 of hfq mRNA) was performed using the pUhfqwt template. Second, a run-off transcript was synthesized in vitro by T7 RNA polymerase. This hfq transcript contained nucleotides −68 to +102. The [³²P]-5′-end-labeled oligonucleotides Y19 or F22 (5′-CACGTTCCCGA C-3′; complementary to nucleotides +56 to +67 of hfq mRNA) were annealed to hfq mRNA (nucleotides −68 to +102) and used to prime cDNA synthesis by reverse transcriptase. The toeprinting assays were carried out with purified 30S ribosomal subunits and initiator-tRNA, tRNA_f^Met, essentially as described by Hartz et al. (1988). The mRNA (0.04 pmol) was pre-incubated at 37°C for 5 min with or without 2 pmol 30S subunits and 10 pmol tRNA_f^Met. To test the effect of Hfq on ternary complex formation, Hfq protein was added prior to the addition of 30S subunits and tRNA_f^Met to the toeprinting reactions at the molar ratios to mRNA as specified in the legend to Figure 2.

Hydroxyl radical footprinting

[³²P]-5′-end labeled hfq126 or hfq96 mRNA (5 nM each) and increasing amounts of Hfq-hexamer (0 nM, 20 nM, 40 nM, and 80 nM) were mixed in the same manner as for the gel shift assay. After a 15 min incubation, the complex was subjected to hydroxyl radical cleavage. The radicals were generated using fresh 1 mM diammonium iron (II) sulfate hexahydrate, 2 mM EDTA, 1 mM sodium ascorbate, and fresh 0.5% hydrogen peroxide (Tullius et al. 1986). The reaction was incubated for 1 min at room temperature and then quenched by addition of 10 mM thiourea. After phenol extraction, the RNA was precipitated and analyzed on a 12% polyacrylamide- 8M urea gel.

RNase T1 probing

Unlabeled hfq transcript (0.1 pmol) containing nucleotides −68 to +102 (see toeprinting experiment) was hybridized to 0.2 pmol [³²P]-5′-end labeled oligonucleotide Y19 by incubation in 50 mM Tris-HCl at pH 8.3, 60 mM NaCl, and 10 mM DTT for 3 min at 85°C and snap freezing in liquid nitrogen. MgCl₂ was added to a final concentration of 2 mM and the incubation was continued for 5 min at 25°C. Then 0.1 or 0.5 units of RNase T1 were added, and the incubation was continued for 5 min at 25°C. The cDNA synthesis was performed in a total volume of 10 μL with 1 U of AMV reverse transcriptase (Promega) and 1 mM of each dNTP for 15 min at 48°C. After primer extension, the reactions were terminated by addition of 10 μL loading buffer and by heating to 95°C. The samples were analyzed on a 10% polyacrylamide-8M urea gel.

In vitro translation

The full length ppiB, hfq, and hfqΔ30 mRNAs were translated in vitro using a S30 extract (Promega) as specified in the manufacturer’s instructions. The translation reactions were incubated for 20 min at 37°C in the presence of [¹⁴C] lysine with or without addition of Hfq as indicated in Figures 2A and 3A. The reactions were terminated by addition of four volumes of cold 90% acetone and placed on ice for 15 min followed by centrifugation at 10,000g at 4°C for 10 min. The pellets were dried, resuspended in 40 μL protein loading buffer, and boiled for 5 min before loading onto a 12% SDS–polyacrylamide gel. The translation products were visualized by autoradiography.