Dynamic mechanisms of CRISPR interference by Escherichia coli CRISPR-Cas3 | Panda Anku

In vitro reconstitution of Escherichia coli CRISPR-Cas3 interference

E. coli CRISPR-Cas3 is generally well-characterized type I CRISPR complexes in vitro and in vivo32,33,37,38. However, recombinant EcoCas3 protein is difficult to purify because of poor solubility and propensity to aggregate at 37 °C25,26,30,39. Co-expression of HtpG chaperon40 and/or low temperature growth at 20 °C24,25 produced only a limited amount of protein that was highly aggregated (Supplementary Fig. 1a). This is in contrast to isolated Thermobifida fusca Cas3 (TfuCas3) protein produced by the E. coli bacterial expression system at 37 °C10,15,31 (Supplementary Fig. 1b). We then used Sf9 insect cells with a baculovirus expression system at 20 °C to produce EcoCas3 protein, which was soluble and 95% homogeneous (Supplementary Fig. 2a, b). EcoCascade proteins and crRNA were co-expressed in E. coli JM109(DE3) and purified using Ni-NTA resin as previously reported24,41 (Supplementary Fig. 2c, d). Tycho NT.6 protein stability measurements (Supplementary Fig. 3) and the ProteoStat protein aggregation assay (Supplementary Fig. 4) indicated that the temperature-dependent stability and aggregation onset temperature of EcoCas3 was consistent with a mesophilic protein25,26,30,39. Then, we confirmed co-purified recombinant EcoCascade-crRNA ribonucleoproteins (RNPs) bound to supercoiled (SC) plasmids composed of hEMX1 spacer sequences flanked by a PAM (5′-AAG-3′), but not bound to SC plasmids including spacer sequences flanked by a nonPAM (5′-CCA-3′) (Supplementary Fig. 5a). Binding of EcoCascade-crRNA RNPs was also observed for linear dsDNA molecules (mTyr and rIl2rg genes) (Supplementary Fig. 5b), and exchanging nucleotide pairs between crRNAs and target sequences abolished this binding (Supplementary Fig. 5c). Finally, assembly of EcoCascade RNPs with purified EcoCas3 protein specifically degraded SC plasmids (hEMX1) and linear dsDNA (mTyr) in the presence of ATP and Mg2+24,25,30,39,42 (Supplementary Fig. 6a, b).

The EcoCas3-EcoCascade-crRNA complex nonspecifically cleaves ssDNAs in trans

Cas3 proteins from Streptococcus thermophiles, Methanocaldococcus jannaschii, and Thermus thermophilus can exhibit indiscriminate, divalent cation-dependent ssDNase activity in the absence of Cascade30,39,42. Using fluorescent dye-quencher (FQ)-labeled ssDNA probes, we found that EcoCas3 and TfuCas3 also exhibit nonspecific ssDNA cleavage in a metal-dependent manner, although the dependency was different between the two bacteria (Fig. 1b). TfuCas3 cleaved ssDNA with all divalent ions tested, whereas EcoCas3 was only activated with Mn2+ and Ni2+, consistent with previous results25. Type V Cas12a, an RNA-guided DNase34, and type VI Cas13, an RNA-guided RNase43, engage in collateral cleavage of nearby non-specific nucleic acids after their targeted activity. To investigate whether Cas3 also possesses collateral ssDNA cleavage activity, we assembled EcoCas3, EcoCascade RNPs, 60 bp dsDNA fragments containing target sequences flanked by a PAM (targeted Activator), and a untargeted ssDNA44. We found that targeted degradation triggered untargeted degradation of both circular M13 phage ssDNA and linearized long ssDNA, but not of circular pBlueScript dsDNA (Supplementary Fig. 7a). As is the case with target dsDNA degradation by EcoCascade RNPs and EcoCas3 (Supplementary Fig. 6a, b), this collateral ssDNA cleavage was dependent upon the presence of a PAM in the targeted nucleic acid (Supplementary Fig. 7a). These results indicate that either some metal ions or Cascade target-binding by R-loop formation can induce EcoCas3-dependent non-specific ssDNA cleavage activity in vitro.

To quantitatively measure collateral ssDNA cleavage activity we used a FQ-labeled untargeted ssDNA probe34,43 (Fig. 1c), which is used in CRISPR-based diagnostics as a platform for rapid and sensitive nucleic acid detection, for example in Covid-19 test kits45,46,47. Consistent with the results of the M13/linear ssDNA cleavage (Supplementary Fig. 7a), EcoCas3 showed collateral ssDNA cleavage in a PAM-dependent manner (with a PAM of 5′-AAG-3′ or 5′-ATG-3′, but not 5′-CCA-3′) (Fig. 1c and Supplementary Fig. 7b, c). Fluorescent reporter DNA oligonucleotides (DNaseAlert™ IDT) also confirmed this collateral cleavage activity (Supplementary Fig. 8a), whereas fluorescent reporter RNA oligonucleotides (RNaseAlert™, IDT) detected little or no collateral RNase activity (Supplementary Fig. 8b). We previously showed that mutants of EcoCas3 in the HD domain (H74A, dead nuclease, dnCas3), and SF2 motif III (S483A/T485A, dead helicase, dhCas3) abolished target DNA degradation in human cells11. In the collateral cleavage assay, the dnCas3 mutant abolished all cleavage activity, but the dhCas3 mutant showed a similar level of activity as that of wild-type EcoCas3 (Fig. 1d). In ATP-free reaction buffer (−), the collateral activity of the EcoCas3 protein was at the same level or higher compared with that of wild-type EcoCas3 and the dhCas3 mutant in ATP (+) buffer (Fig. 1d). Together, these results indicate that trans cleavage of non-specific ssDNA is catalyzed by the single HD domain of EcoCas3 and that the helicase activity is not required for the trans cleavage but may restrain it.

PAM recognition is a prerequisite for collateral ssDNA cleavage by Cas3 but not Cas12a

Having determined that collateral ssDNA cleavage by the EcoCas3-EcoCascade complex is PAM-dependent (Fig. 1c), we sought to further characterize the specificity of PAM recognition by screening all 64 possible target sites containing each of the three-nucleotide PAM sequences (Fig. 2a and Supplementary Fig. 9a). We observed collateral cleavage activity with 14 PAM types, with the highest activity from 5′-AAG-3′ and 5′-ATG-3′, followed by 5′-GAG-3′, 5′-AAA-3′, 5′-AAC-3′, 5′-TAG-3′, and 5′-AGG-3′. There was no cleavage when the first or second PAM nucleotide was C or the third nucleotide was T (Fig. 2a and Supplementary Fig. 9a). This PAM recognition specificity for trans cleavage activity matched the results from an in vivo high-throughput CRISPR-interference assay48. In contrast, LbaCas12a showed collateral cleavage activity with almost all 64 PAM types, with the highest activity with 5′-GGGG-3′ and the lowest with 5′-GCCG-3′ (Fig. 2b and Supplementary Fig. 9b).

Fig. 2: Specificity of PAM recognition for collateral ssDNA cleavage by EcoCas3.
figure 2

Screening of all 64 possible target sites containing each of the three-nucleotide PAM sequences for trans cleavage activity by EcoCas3 (a) and by LbaCas12a (b). The heat maps represent the RFU per min for collateral cleavage activity. c Trans ssDNA cleavage by a crRNA-complementary or non-complementary ssDNA (TS or NTS, respectively). EcoCas3/EcoCascade partially activated by TS ssDNA in a PAM-dependent manner (3′-TTC-5′ only). d LbaCas12a activated by TS ssDNA in a PAM-independent manner (both 3′-AAAC-5′ and 3′-TTTG-5′). e Screening of all 64 possible target sites containing each of the three-nucleotide PAM sequences for collateral cleavage activity by the TS ssDNA. f Collateral cleavage activated by crRNA-complementary ssDNA, where the PAM region was double stranded but the rest of the protospacer was single stranded. dsDNA-activated collateral cleavage is also shown to the right. Row data in the heatmaps of a, b, e is available in the Source data file. Data in c, f are presented for n = 3 independent measurements and mean value, error bars represent SD values. Data in d are presented for n = 2 independent experiments with central value.

According to previous reports34,35,36, binding of the ssDNA complementary to the crRNA activates Cas12a for nonspecific trans cleavage. We also observed that EcoCas3 and LbaCas12a were activated by crRNA-complementary ssDNA (TS) but not by non-complementary ssDNA (NTS) (Fig. 2c, d). However, the PAM specificity was different between EcoCas3 and LbaCas12a. LbaCas12a was activated by both crRNA-complementary TS flanked by a PAM (3′-AAAC-5′) or a nonPAM (3′-TTTG-5′) (Fig. 2d), as previously reported34,35. In contrast, EcoCas3 was partially activated by a TS with a PAM (3′-TTC-5′) but a TS with a nonPAM (3′-GGT-5′) prevented any activity (Fig. 2c). One potential explanation for why the PAM specificity was not observed in LbaCas12a is that the dsDNA targets are not fully double stranded, and some ssDNA contamination is sufficient to activate LbCas12a. We then tested TS PAM specificity for all 64 possible target sites (Fig. 2e). The PAM specificities for ssDNA-activated collateral cleavage were similar to those of dsDNA-activated collateral cleavage (Fig. 2a), although the activity was mostly lower for ssDNA-activated cleavages, except for when the third nucleotide of the PAM was C, such as 3′-TAC-5′, 3′-AGC-5′, 3′-GTC-5′, 3′-GAC-5′, and 3′-GGC-5′, when the relative fluorescence was increased (Fig. 2e). This means the third nucleotide of the TS PAM is important for the collateral cleavage.

Base-pairing between the TS and NTS of the PAM leads to correct Cascade/crRNA binding of the NTS, accessibility of the EcoCas3 cleavage site, and degradation of the target DNA24,49. We observed that dsDNA containing an unpaired PAM between NTS-nonPAM (5′-CCA-3′) and TS-PAM (3′-TTC-5′) partially activated EcoCas3 for collateral ssDNA cleavage (Supplementary Fig. 10a). This is in contrast to previous reports of dsDNA with an unpaired PAM not showing any activity for target dsDNA degradation24,49. Screening of PAM base-pairing between each of the three nucleotides showed that base-pairing of the third nucleotide positively affected collateral cleavage activity, and that base-pairing of the first and second nucleotides additively increased the activity of the third nucleotide base-pairing (Supplementary Fig. 10b). Interestingly, the crRNA-complementary ssDNA, where the PAM region was double stranded but the rest of the protospacer was single stranded, displayed higher activity than either ssDNA- or dsDNA-activated collateral cleavage (Fig. 2f). This is probably because the protospacer ssDNA can easily bind to the complementary crRNA without unwinding the target dsDNA, which is followed by Cas3 recruitment and activation for ssDNA cleavage. Together, these results of PAM recognition specificity are mostly consistent with results from in vitro reconstitution24,49 and of crystal structure analysis18, except for the partial activity detected for collateral cleavage, in contrast to no activity for target DNA degradation by unpaired PAM recognition24.

EcoCas3 cleaves the NTS in cis followed by the TS in trans in a helicase-dependent manner

Complete R-loop formation by the Cascade/crRNA complex recruits the Cas3 helicase/nuclease, which repeatedly cleaves the NTS via the HD domain’s single catalytic site32,33. It remains unknown how EcoCas3 cleaves the TS and progressively degrades the dsDNA substrate (Fig. 1a). Considering the collateral non-specific ssDNA cleavage in trans, we hypothesized that the TS can be cleaved in trans, following cis cleavage of the NTS after target dsDNA unwinding by the helicase properties of Cas3. To test this, we designed fluorescently labeled target dsDNA substrates, 5′-NTS-FAM, and 5′-TS-TAMRA, to visualize dsDNA cleavage by EcoCas3 (Supplementary Fig. 11a). In control experiments, SpCas9 cleaved both NTS and TS at 3–4 nucleotides upstream of the PAM site, as expected (Fig. 3a). In contrast, the highest peak of EcoCas3 cleavage was 10–11 nucleotides downstream of the PAM site on the NTS, while several peaks upstream of the PAM site demonstrated repetitive cleavage of the NTS. We also observed repetitive cleavage of dozens of nucleotides upstream of the TS PAM, which was likely reeled by EcoCas3 helicase activity and cleaved by its trans cleavage activity (Fig. 3a).

Fig. 3: Mechanistic insight into collateral ssDNA cleavage and target DNA degradation.
figure 3

a Fluorescently-labeled target dsDNA substrates, 5′-NTS-FAM, and 5′-TS-TAMRA, to visualize dsDNA cleavage. EcoCas3 with EcoCascade RNPs cleaves at NTS nucleotides 10–11, downstream of the PAM site, with repetitive cleavage. The TS cleaved repetitively dozens of nucleotides upstream of the PAM. SpCas9 cleaves both NTS and TS at 3–4 nucleotides from the PAM. The x-axis represents the DNA fragment size in base pairs (bp), and the y-axis represents the arbitrary unit (a.u.). b The dnCas3 HD domain mutant and the dhCas3 SF2 domain mutant for the dsDNA cleavage assay. c The dsDNA cleavage assay by EcoCas3 in ATP (+) or ATP-free (−) reaction buffer. d Changing the size of the R-loop formation from 0 to 32 nucleotides by adding multiples of six nucleotides. e Measurement of EcoCascade-target DNA associations and dissociations in real-time using a bio-layer interferometry (BLI) biosensor (Octet RED 96 system). f In the collateral cleavage assay, n0–n12 base-pair hybridization did not show any cleavage activity, while n18–n32 R-loop formations increasingly promoted trans ssDNA cleavage activity. Data are presented for n = 3 independent measurements and mean value, error bars represent SD values. g In the dsDNA cleavage assay, the n0–n24 R-loop formation did not produce any cleavage of the NTS or TS. The 30–32 base-pair R-loop formations underwent repetitive cleavage on the NTS and the TS of the target dsDNA substrates. Source data are provided as a Source Data file.

To confirm the NTS and TS cleavages mediated by nuclease/helicase activities of EcoCas3, we tested a dnCas3 HD domain mutant and a dhCas3 SF2 domain mutant in the dsDNA cleavage assay (Fig. 3b). The dnCas3 mutant cleaved neither NTS nor TS, indicating that the single catalytic domain of EcoCas3 plays a role in generating double-strand breaks (DSBs). Notably, the dhCas3 mutant cleaved the NTS, but not the TS, indicating that the dhCas3 mutant (S483A/T485A) works as an EcoCas3 Nickase (Fig. 3b). In ATP-free reaction buffer, the wild-type EcoCas3 and the dhCas3 mutant also cleaved the NTS, but not the TS (Fig. 3c and Supplementary Fig. 11b). Given the assay’s collateral cleavage results, where the dhCas3 mutant cleaved non-specific ssDNA (Fig. 1f), the helicase activity of EcoCas3 followed by reeling of TS is indispensable not only for repetitive cis cleavage of the NTS but also for trans cleavage of the reeled TS.

To further characterize cis and trans cleavage by EcoCas3, we compared 30 s (short) and 5 min (long) incubation times for the dsDNA cleavage assay. More prolonged incubation increased repetitive cleavage of the NTS in cis and the TS in trans (Supplementary Fig. 12a). We also observed that progressive cis and trans cleavages showed similar patterns in the repetitive experiments and the short and long incubation experiments, depending on the target DNA sequence, e.g., hEMX1 and mTyr (Supplementary Fig. 12a). The sizes of many cleaved fragments were between 30 and 60 bps, which may be used for CRISPR adaptations as previously reported12,50 (Supplementary Fig. 12b).

Incomplete binding of EcoCascade to target DNA with tolerated mismatches elicits collateral ssDNA cleavage but not target dsDNA degradation

We previously reported that a single mismatch within the seed region markedly affected target DNA degradation in the EcoCascade/Cas3 system11. We, therefore, investigated the effect of mismatch for each nucleotide in the 32-nt spacer on collateral ssDNA cleavage activity. A single mismatch in the spacer region, even within the seed region (positions 1–8), resulted in little or no effect on collateral cleavage activity (Supplementary Fig. 13a, b). In the LbaCas12a system, 1–3 mismatches in the seed region also did not affect collateral cleavage activity (Supplementary Fig. 13c), consistent with previous reports36,51. Previous in vitro analysis revealed the effect of single mismatches in the target sequence, which slow the rate of R-loop formation and target-strand cleavage by Cas12a52,53. To investigate whether Cascade-binding and R-loop-formation are linked with collateral cleavage and target DNA degradation, we sought to characterize Cascade-target DNA binding kinetics using a Bio-layer interferometry (BLI) biosensor54. Corresponding to the collateral cleavage assay results (Fig. 2c), crRNA-complementary TS-ssDNA showed associations with EcoCascade but not with non-complementary NTS-ssDNA (Supplementary Fig. 14a and Supplementary Table 1). Notably, the crRNA-complementary TS-PAM (3′-TTC-5′) showed higher association than that of TS-nonPAM (3′-GGT-5′) or -PAMless (Supplementary Fig. 14a). Moreover, dsDNAs containing a paired PAM (5′-AAG-3′ − 3′-TTC-5′) showed the maximum EcoCascade-target DNA binding (Supplementary Fig. 14b and Supplementary Table 1), which corresponds to the results of the collateral cleavage assay (Fig. 2f). Unpaired PAM between TS-PAM (3′-TTC-5′) and NTS-nonPAM (5′-CCA-3′) indicated a lower association, and unpaired PAM between NTS-PAM (5′-AAG-3′) and TS-nonPAM (3′-GGT-5′) showed little association (Supplementary Fig. 14b). Taken together, BLI can provide solid information on the affinity and stability of interactions as previously reported54.

To further investigate the relationship between the R-loop-formation and EcoCas3-mediated collateral ssDNA cleavage and dsDNA degradation, we assayed different length R-loop formations, from 0 to 32 nucleotides (n0, n6, n12, n18, n24, n30, and n32) (Fig. 3d). BLI revealed that crRNA-DNA hybridization with 0–12 base-pairs (n0, n6, and n12) including seed sequences did not show any association, while 18–30 base-pairs (n18, n24, and n30) produced a degree of association. Furthermore, a complete match for 32 base-pairs (n32) resulted in stable and emphatic Cascade binding, similar to locked R-loop formation reported previously15,19,29 (Fig. 3e and Supplementary Table 1). In the collateral cleavage assay, n0, n6, and n12 did not show any cleavage activity, while n18–n32 R-loop formations increasingly promoted trans cleavage of ssDNA (Fig. 3f). In the dsDNA cleavage assay, n0–n24 did not show any cleavage of either NTS or TS DNA (Fig. 3g). This means that collateral cleavage does not need the nicking activity on the NTS (n18 and n24). Only the n30 and n32 sequences underwent repetitive cleavage on both the NTS and TS, and progressive cleavage of target dsDNA substrates (Fig. 3g). Taken together, these results show two Cascade binding modes. Intermediate R-loop formation by mismatches on the spacer sequences elicits collateral ssDNA cleavage. Complete R-loop formation with full crRNA-DNA hybridization leads to repetitive cis cleavage of the NTS with trans cleavage of the TS to degrade the target dsDNA substrate, as described in previous reports15,19,29.

Dynamic visualization of CRISPR interference: PAM search, nicking, and DSB

Cryo-EM and smFRET are not capable of visualizing how EcoCas3 degrades target dsDNA12,31 (Fig. 1a). We therefore employed hs-AFM, which enables real-space and real-time observations of CRISPR-Cas3 reacting with target DNAs at the macromolecule level, as previously shown by visualizing CRISPR-Cas9 interference55. First, we visualized the binding of Cascade/crRNA to a target DNA, a 645 bp dsDNA containing a target spacer site flanked by a PAM (5′-AAG-3′) at 219 and 423 bp from the ends of the DNA fragment (Fig. 4a). We adsorbed the mixture of donor DNAs and EcoCascade RNPs onto a 3-aminopropyltrietoxysilane-mica surface (APTES-mica)56. As expected, hs-AFM demonstrated that EcoCascade RNP specifically bound to the expected target site in the dsDNA (Supplementary Fig. 15a–d). Notably, we frequently observed a typical DNA bend at the EcoCascade-RNP binding site (Fig. 4b). The DNA angle distribution around the bound EcoCascade RNP exhibited a large spread with a peak near 45°, while the DNA angle distribution around bare DNA showed an intrinsic gradual curve reflecting the persistence length of dsDNA, with or without Cas357,58 (Fig. 4c, d and Supplementary Fig. 15e, f). This DNA bending by a Cascade for stable R-loop formation was previously indicated by cryo-EM15,18 and smFRET studies16,59. During the observation periods, the EcoCascade RNPs bound tightly to the target DNAs without dissociating, consistent with previous smFRET analyses16,20,59.

Fig. 4: Dynamic visualization of CRISPR-Cas3 binding and nicking by hs-AFM.
figure 4

a Schematic depictions of Cascade binding and Cas3-mediated nicking. b hs-AFM visualizes the EcoCascade RNP (green open triangle) searching for an appropriate PAM site from one end of the target DNA to the other (green arrows) (Video 1). EcoCascade binds to the target DNA (green closed triangle), a 674 bp dsDNA containing a target spacer site flanked by PAMs (5′-AAG-3′) at 219 bp and 423 bp from the ends of the DNA fragment (purple and blue triangle, respectively). Scan area: 200 × 200 nm2 with 80 × 80 pixels; frame time: 0.25 s (line rate: 320 Hz). c Overview of strategies for measuring angles around Cascade. R:15 nm. d Histograms of DNA angles around P1 (top) and around P2 (bottom) in EcoCascade-dsDNA. e Injection of EcoCas3 protein after immobilizing EcoCascade RNPs with the target DNA in ATP-free (−) reaction buffer to produce EcoCas3-mediated nicking (white triangle) at the target site (video 3). Scan area: 150 × 150 nm2 (left 4 images) and 100 × 100 nm2 (the most right image) with 80 × 80 pixels; frame time: 0.2 s (line rate: 400 Hz). The experiment of b, e was repeated twice independently with similar results.

We also observed target interrogation by the EcoCascade RNP on a mica-supported lipid bilayer (mica-SLB), which was indispensable to observe the dynamic events of EcoCascade RNPs, as previously used in the CRISPR-Cas9 system55. While the Cascades bound to the target site for more than 10 s in all cases, the duration of the non-target Cascade was mostly measured to be less than 1 s (Supplementary Fig. 16). We sometimes observed that the EcoCascade RNP ran from one end to the other through the target DNA (Fig. 4a, b and Videos 1 and 2), presumably searching for the right PAM site and spacer sequences, as shown by the smFRET studies16,59.

Next, we injected EcoCas3 proteins after immobilizing the EcoCascade RNPs with the 645-bp target DNA in ATP-free reaction buffer on the mica-SLB to reproduce EcoCas3-mediated nicking at the target site. In this preparation, several single-strand break (SSB)-like DNAs at the Cascade binding site were observed (Fig. 4e and Video 3). We measured the DNA height because the preliminary experiments revealed a decrease in height at the nick site in artificially nicked DNA using Nb.BsrDI nicking endonucleases on APTES mica (Supplementary Fig. 17). Therefore, we observed pre-mixed EcoCas3-EcoCascade-dsDNA on APTES mica and measured the height of DNA for a more quantitative analysis. When removing the protein complex from dsDNA by applying extensive force, we observed a nick-like shape, where the DNA chain remained connected, but the height of dsDNA appeared to be partially lower in 60% of the complex (Supplementary Fig. 18a, b and Videos 3 and 4). In contrast, no nick structure was observed after mechanically removing EcoCascade from dsDNA without EcoCas3 protein (Supplementary Fig. 18c–e). These observations, therefore, demonstrate that EcoCas3 induces the nick as depicted in the in vitro experiments in Fig. 3.

In contrast, in ATP-containing reaction buffer, we detected many short DNA fragments after injection of EcoCas3 proteins (Supplementary Fig. 19). The DNA fragment lengths obtained varied widely but were commonly approximately 200 bps. Notably, we observed the EcoCas3-Cascade complex bound to the target site was repeatedly reeling the longer side of the DNA then releasing it, as previously reported20 (Fig. 5a, b, Supplementary Fig. 20 and Videos 5–7). Finally, we captured the dynamic movements by which the EcoCas3-Cascade complex shortened the target DNA and cleaved it with a DSB after the reeling reaction, followed by release of the DNA from the EcoCas3-Cascade complex (Fig. 5b and Videos 6 and 7). These results clearly indicated that the DNA fragments can be cleaved upstream of the target site when they are shortened by reeling, rather than by Cas3 translocation.

Fig. 5: Visualization of type I CRISPR interference.
figure 5

a Schematic depictions of type I CRISPR interference. b In ATP (+) reaction buffer, the EcoCas3-Cascade complex repeatedly reels and releases the longer side of the DNA (blue arrows) and then cleaves it with a DSB (red arrows) (Video 5). Scan area: 200 × 200 nm2 with 80 × 80 pixels; frame time: 0.2 s (line rate: 400 Hz). This experiment was repeated twice independently with similar results. Source data are provided as a Source Data file.

Leave a Comment