NsrR-dependent method for detecting nitric oxide accumulation in the Escherichia coli cytoplasm and enzymes involved in NO production


Correspondence: Jeffrey A. Cole, School of Biological sciences, University of Birmingham, Birmingham B15 2TT, UK. Tel.: +44 121 414 5440; fax: +44 121 414 5925; e-mail: j.a.cole@bham.ac.uk


A β-galactosidase assay for detecting the accumulation of NO in the Escherichia coli cytoplasm has been developed based on the sensitive response of the transcription repressor, NsrR, to NO. The hcp promoter is repressed by NsrR in the absence of nitric oxide, but repression is relieved when NO accumulates in the cytoplasm. Most, but not all, of this NO is formed by the interaction of the membrane-associated nitrate reductase, NarG, with nitrite. External NO at physiologically relevant concentrations does not equilibrate across the E. coli membrane with NsrR in the cytoplasm. The periplasmic nitrite reductase, NrfAB, is not required to prevent equilibration of NO across the membrane. External NO supplied at the highest concentration reported to occur in vivo does not damage FNR sufficiently to affect transcription from the hcp or hmp promoters or from a synthetic promoter. We suggest that the capacity of E. coli to reduce NO is sufficient to prevent its accumulation from external sources in the cytoplasm.


The damaging effects of nitric oxide on proteins, lipids and DNA are well established. Bacteria are exposed to reactive nitrogen species generated from nitrate or nitrite in their environment, generated externally from arginine as a part of the nitrosative burst of mammalian host defence mechanisms, or as products of nitrate, nitrite or ammonia metabolism by bacteria that share their immediate environment. Enteric bacteria have developed multiple mechanisms for protecting themselves from reactive nitrogen species, such as nitric oxide.

Smith (1983) demonstrated that small quantities of NO are generated during nitrite reduction, but the rate of NO generation was estimated to be two to three orders of magnitude less than the maximum rate of nitrite reduction to ammonia. In an anaerobic environment, Escherichia coli reduces nitrite rapidly to ammonia using either of two pathways. There is a cytoplasmic, NADH-dependent nitrite reductase, NirBD, that is synthesized in response to the availability of high concentrations of nitrate. The alternative nitrite reductase, NrfAB, is located in the periplasm and is preferentially synthesized in response to the availability of low concentrations of nitrate. It was largely assumed that NO is a side product released during nitrite reduction by one or both of these nitrite reductases. Although there are experimental data to support this suggestion (Corker & Roole, 2003; Weiss, 2006), other studies with both E. coli and Salmonella enterica have implicated the nitrate reductase, NarG, as the enzyme that generates most of the NO when nitrite is abundant, but nitrate is unavailable (Calmels et al., 1988; Ralt et al., 1988; Metheringham & Cole, 1997; Gilberthorpe & Poole, 2008). Recently, it has been realized that five or more proteins catalyse the reduction of either NO itself or NO attached to nitrosylated proteins or S-nitrosoglutathione. These include flavorubredoxin and its reductase (NorV-NorW), flavohaemoglobin (Hmp), cytochrome c nitrite reductase (NrfA), S-nitrosogluathione reductase, AdhC and possibly also the cytoplasmic nitrite reductase, NirBD. Considerable doubt remains about the concentration of NO that accumulates inside enteric bacteria, its physiological consequences and how rapidly cytoplasmic NO is generated or removed.

Spiro (2007) has emphasized the need to distinguish between direct effects of physiological concentrations of NO on gene regulation, and secondary effects due to chemical damage to iron-sulphur centres of transcription factors caused by higher concentrations of NO. Bacteria rarely, if ever, encounter NO at concentrations above 1 μM, the exception being intracellular bacteria, such as S. enterica in macrophages, where the concentration of NO has been estimated to be up to 10 μM (Raines et al., 2006). As NO is an uncharged small molecule that is freely diffusible across membranes, it is assumed that NO generated by the host will equilibrate with the bacterial cytoplasm. We have found no direct evidence in the literature that this assumption is correct.

A previously described method for detecting the accumulation of NO in the cytoplasm was based on the heterologous expression in E. coli of the NO-sensitive transcription factor, NNR, from Paracoccus denitrificans and its ability to activate transcription from an engineered E. coli melR promoter (Hutchings et al., 2000). A similar principle was used by Cruz-Ramos et al. (2002) to detect NO-induced damage to the transcription factor, FNR, and by Strube et al. (2007) to detect availability of free NO in the cytoplasm of Ralstonia eutropha to detect transcription activation of the norA promoter by NorR. This led us to develop an assay based on the fact that the transcription factor, NsrR, responds specifically and with very high sensitivity to NO located in the cytoplasm rather than outside the cytoplasmic membrane (Bodenmiller & Spiro, 2006; Tucker et al., 2008). The assay was used to compare the effects on NsrR-dependent transcription of mutations in genes for enzymes implicated in NO production, as well as the effectiveness of externally added NO and nitrite as sources of cytoplasmic NO.

Materials and methods

Strains, plasmids and growth conditions

Strains of E. coli K-12 and plasmids used in this study are listed in Table 1. The nsrR::kan mutation was transferred by P1 transduction from E. coli strain JOEY 60 to RK4353 to construct strain JCB 5222. The strain to be tested was transformed with the Phcp::lacZ fusion plasmid, pNF383, (Filenko et al., 2007). A plasmid with a synthetic promoter with a consensus FNR-binding site linked to lacZ that is repressed by FNR was used in control experiments designed to distinguish between NO-induced damage to FNR and NO-induced derepression of Phcp (Williams et al., 1998). Purified transformants were grown in minimal salts medium (MS) supplemented with 5% (v/v) LB, 0.4% (v/v) glycerol, 20 mM trimethylamine-N-oxide, 20 mM sodium fumarate and 35 μg mL−1 tetracycline. Cultures were started with 2% inocula that had been grown overnight at 37 °C with aeration in 5 mL LB in 25 mL conical flasks. Multiple anaerobic cultures were incubated statically at 37 °C in test tubes filled with 15 mL of medium. Once the optical density at 650 nm had reached 0.2 or above, one culture was left as an unsupplemented control; other cultures were supplemented as stated in the text with 2.5 or 10 mM sodium nitrite, 20 mM sodium nitrate, or 5–20 μM nitric oxide saturated water (NOSW) prepared as described by Vine & Cole (2011). Nitric oxide saturated water was added repeatedly at 30 min intervals under the surface of the culture using a sterile syringe and needle to avoid exposure to oxygen. A magnetic stirrer was used very briefly to ensure that the NOSW was distributed evenly throughout the culture, but to avoid aeration. Cultures were incubated statically at 37 °C.

Table 1. Escherichia coliK-12 strains and plasmids used in this study
JCB 387RV nir lacPage et al. (1990)
JCB 3911RV nir lac fnrSquire et al. (2009)
JCB 4022RK4353 ΔnarG::ery ΔnapA-BPotter et al. (1999)
JCB 4031narG-I narZ derivative of strain RK4353Potter et al. (1999)
JCB 5205nirBDC derivative of strain RK4353Vine & Cole (2011)
JCB 5206nrfAB derivative of strain RK4353Vine & Cole (2011)
JCB 5222nsrR derivative of strain RK4353This work
JOEY 60araD139 (ara-leu) (codB-lacI) galK16 galE15 relA1 rpsL spoT1 nsrR::kanBodenmiller & Spiro (2006)
RK4353Parent strainStewart & McGregor (1982)
FF gal Δ4A synthetic FNR-repressed promoter fused to lacZ in pRW50Williams et al. (1998)
pNF383The hcp regulatory region cloned into pRW50 to give an hcp::lacZ transcriptional fusion. TetRFilenko et al. (2007)
pRW50Promoter-probe lacZ fusion vector for cloning promoters as EcoRI-HindIII fragmentsLodge et al. (1992)
pSP01The hmp regulatory region cloned into pRW50 to give an hmp::lacZ fusion. TetRThis work

Growth and preparation of bacteria for use in the nitric oxide electrode

Bacteria were grown in MS supplemented where indicated with 10% LB, 0.4% glycerol, 20 mM TMAO, 20 mM sodium fumarate and 2.5 mM sodium nitrite or 20 mM sodium nitrate. All cultures were started with inocula that had been grown at 37 °C with aeration in 2 mL LB in a test tube for at least 2 h. Anaerobic cultures were incubated statically overnight at 30 °C 250 mL flasks filled with 250 mL of medium. The optical density at 650 nm was monitored until it had reached 0.6–0.8, then the bacteria were collected by centrifugation (8000 g, 2 min, 4 °C). The bacteria were resuspended in 10 mL phosphate buffer and were homogenized. The washed bacteria were collected by centrifugation (3000 g, 3 min), then resuspended in 0.5–1 mL phosphate buffer to give an optical density at 650 nm of 70–90. The bacteria were kept on ice until required.

Preparation of nitric oxide saturated water and NO electrode assays

An Oxytherm electrode control unit was used with an S1/MINI Clark type electrode disc and the Oxygraph Plus data acquisition software (Hansatech Instruments, Norfolk, UK) to measure the rate of NO reduction. To prepare NO-saturated water, 5 mL of distilled water in a glass bijoux bottle that was sealed with an airtight septum (Fisher Scientific, Leicestershire, UK) was flushed for 30 min with nitrogen that had been passed through sealed bottles of distilled water and 3 M NaOH, then with nitric oxide for 30 min. The needles were removed and the bottles were sealed with Parafilm to prevent any oxygen leaking into the vessel. When required, NOSW was removed from the bottles using an airtight syringe. The assay buffer was also degassed with nitrogen in the same way. Reagents were added to the electrode chamber using Gastight High-Performance syringes (Hamilton, Bonaduz, Switzerland). The assay buffer was 50 mM sodium phosphate, pH 7.5, supplemented with 50 μM EDTA and 0.4% v/v glycerol. To calibrate the electrode, 1788 μL degassed assay buffer, 32 μL 1 M glucose, 20 μL glucose oxidase (0.4 U μL−1) and 10 μL catalase (4 U μL−1) were added to the electrode chamber. When the last traces of oxygen, which is also detected by the electrode, had been removed, 150 μL of 2 mM NO-saturated water was added and the trace was checked carefully to ensure that the reading was in the range expected (50–150 units) and that there was no rate of NO reduction in the absence of bacteria. The amplitude of the electrode response was noted. To assay rates of NO reduction by bacteria, 1688 μL of degassed assay buffer, 32 μL of 1 M glucose, 20 μL of glucose oxidase (0.4 U μL−1), 10 μL catalase (4 U μL−1) and 100 μL of bacterial suspension were added to the electrode chamber. The reaction was started by the addition of 25–150 μL of NO-saturated water. The initial rate of NO reduction was then calculated. Using this assay, NO reduction rates were proportional to the concentration of bacteria added, providing the [NO] in the reaction vessel was below 200 μM.

β-galactosidase assays and reproducibility of the data

A 2-mL sample of the culture to be assayed was lysed by incubation for 10 min at 37 °C with 30 μL of a 1% aqueous solution of sodium deoxycholate and 30 μL of toluene. The β-galactosidase activity was determined as described by Jayaraman et al. (1988). Activities are expressed as nmol of orthonitrophenol formed min−1 (mg bacterial dry mass)−1, assuming that an optical density of 1.0 at 650 nm (A650 nm) corresponds to 0.4 g dry mass L−1. Corrections were applied to all assays for the turbidity of the lysed bacteria by subtracting the A420 of samples incubated in the absence of substrate, o-nitrophenol-β-d-galactose, from the absorbance generated in the presence of substrate.

Results reported are representative of at least two biological replicates and at least two assay replicates. However, many of the experiments were repeated five or more times. The error bars shown are the standard deviation of all assays.

Determination of nitrite concentration

Culture samples were centrifuged at 13 000 g for 2 min, and the concentration of nitrite in the supernatant was assayed as described by Pope & Cole (1984).


Derepression of transcription at an NsrR-dependent promoter during growth in the presence of nitrate, nitrite or nitric oxide

As a reporter system, we used a plasmid from which β-galactosidase synthesis is dependent upon relief of NsrR repression of the promoter of the hcp gene in response to cytoplasmic NO (Filenko et al., 2007; Chismon et al., 2010). In this and other earlier work, nitrite was used as a source of NO to study NsrR repression at a range of promoters (Kim et al., 2003; Constantinidou et al., 2006; Vine & Cole, 2011). In initial experiments, duplicate cultures of E. coli strain RK4353 transformed with the Phcp::lacZ fusion plasmid were grown anaerobically to mid-exponential phase in the absence of nitrite, and 2.5 mM nitrite was then added to one culture. Transcription from Phcp was strongly activated in the presence of nitrite, but not in its absence (Table 2).

Table 2. Effects of growth conditions and host genotype on transcription at the hcp and hmp promoters and at a synthetic FNR-repressed promoter
StrainRelevant genotypeReporterInduceraT (min)bβ-galactosidase activityRatio
No inductionWith inducer
  1. a

    Time after inducer addition that sample was assayed. Nitrate or nitrite was added once: sequential additions of NO were made every 30 min.

  2. b

    Units of β-galactosidase activity are nmol of ONPG hydrolysed min−1 (mg dry weight)−1. Error noted is the standard error of all assays.

RK4353ParentPhcp::lacZ 2.5 mM nitrite902750 ± 569150 ± 2403.3
JCB 5222nsrRPhcp::lacZ 2.5 mM nitrite9042 230 ± 7346 180 ± 9101.1
RK4353ParentPhcp::lacZ20 μM NO903040 ± 542405 ± 430.79
JCB 5206nrfABPhcp::lacZ20 μM NO902590 ± 622610 ± 461.0
RK4353ParentPhcp::lacZ 2.5 mM nitrite606515 ± 35716 000 ± 3022.4
JCB 5205nirBDCPhcp::lacZ 2.5 mM nitrite606040 ± 522 100 ± 593.7
JCB 5206nrfABPhcp::lacZ 2.5 mM nitrite605290 ± 2721 900 ± 804.1
JCB 4031narGHJI narZPhcp::lacZ 2.5 mM nitrite605470 ± 2710 100 ± 311.8
JCB 4031narGHJI narZPhcp::lacZ10 mM nitrite1205770 ± 8713 700 ± 3332.4
JCB 4022narG napPhcp::lacZ10 mM nitrite1204220 ± 608030 ± 1701.9
JCB 387ParentPhmp::lacZ 5 μM NO601220 ± 421530 ± 201.2
JCB 3911fnrPhmp::lacZ 5 μM NO601320 ± 661815 ± 201.4
JCB 387ParentFF-37.5::lacZ 5 μM NO120490 ± 5365 ± 4.50.75
JCB 3911fnrFF-37.5::lacZ 5 μM NO1201700 ± 71680 ± 500.99

The experiments were then repeated using an nsrR mutant as the host strain. As expected, high activities were detected both in the presence and absence of nitrite (Table 2). This was consistent with the expectation that the response of the NsrR+ culture to nitrite was dependent upon inactivation of the repressor activity of NsrR by NO that had accumulated in the cytoplasm. However, the response to NO generated from nitrite was far smaller than the effect of an nsrR deletion mutation.

Various sources of NO have been used in different laboratories to study its effects on gene regulation and metabolism, mainly because NO reacts rapidly with oxygen in aerobic cultures. In the absence of oxygen, NO is stable, and so it is possible to avoid using S-nitrosoglutathione, nitrite or other sources of NO as a surrogate for NO. First, an NO-sensitive electrode was used to confirm that NO was stable in the absence of bacteria for long periods under conditions used for subsequent experiments. The effect on bacterial growth of sequential additions of various concentrations of NO at 30 min intervals was then determined (Fig. 1). Growth was totally inhibited at concentrations above 10 μM NO, which is well above the range encountered by bacteria in vivo, and was also inhibited by sequential additions of 5 μM (not shown) or 10 μM NO, but not by 1 μM NO (not shown). As NsrR responds to sub-μM concentrations of NO, 5–20 μM NO was used in subsequent experiments.

Figure 1.

Effect of nitric oxide concentration on Escherichia coli growth. Replicate cultures of E. coli strain RK4353 were grown anaerobically for 2 h and then supplemented every 30 min with nitric oxide saturated water to give the NO concentrations shown. The optical density of samples removed at intervals was determined.

To determine optimal growth conditions for transcription activation at Phcp, further cultures were supplemented with either 10 mM sodium nitrite, 20 mM sodium nitrate, or oxygen-free, NO-saturated water added to a final concentration of 10 μM. Transcription of hcp::lacZ was induced far less by nitrate than by nitrite, but there was even less response to externally added NO, even when supplementation with NO was repeated at 30 min intervals (Fig. 2). These results not only implicated NO production from nitrite as the source of cytoplasmic NO, but also indicated that externally added NO might not equilibrate with the E. coli cytoplasm, possibly because it is reduced as it crosses the periplasm or cytoplasmic membrane.

Figure 2.

Derepression of transcription at the hcp promoter during anaerobic growth in the presence of nitrate, nitrite or nitric oxide. The parental strain RK4353 was transformed with pNF383 and purified transformants were grown anaerobically until the optical density at 650 nm had reached 0.2. Cultures were either unsupplemented (▲), or supplemented once with 10 mM sodium nitrite (●), with 20 mM sodium nitrate (Δ) or at 30 min intervals with 10 μM NO-saturated water (○). Samples were lysed and assayed for β-galactosidase activity just before supplement addition, and at 30 min intervals for 2 h. Units of β-galactosidase activity are nmol ONPG hydrolysed min−1 (mg dry weight)−1.

To estimate the maximum possible rate of NO generation from nitrite, the residual rate of nitrite reduction by a strain defective in both of the E. coli nitrite reductases, NrfA and NirB, was determined after anaerobic growth in the presence of nitrate. This rate was between 1 and 2 nmol of nitrite reduced min−1 (mg of bacterial dry mass)−1. This was an order of magnitude less than the rate of NO reduction by this strain measured using an NO-sensitive electrode, which was 15 or 25 nmol of NO reduced min−1 (mg of bacterial dry mass)−1, depending on whether the bacteria had been grown in the presence of nitrite or nitrate (see also Vine & Cole, 2011).

Effect of a mutation in the periplasmic nitrite reductase on transcription activation by NO

One possible explanation why externally added NO did not induce Phcp::lacZ transcription was that it is reduced by an active NO reductase located either in the periplasm or in the cytoplasmic membrane. An obvious candidate for such NO reductase activity is NrfAB, which despite its high Km for NO has been proposed to fulfil this role with high catalytic efficiency (Poock et al., 2002; van Wonderen et al., 2008). Cultures of the parent strain and the nrfA mutant were therefore supplemented every 30 min with NO to a final concentration of 20 μM and compared with unsupplemented control cultures (Table 1). Loss of NrfAB function did not increase the transcription response of Phcp to NO, indicating that NrfAB is not the enzyme responsible for elimination of externally added NO.

Enzymes implicated in NO production in the E. coli cytoplasm

The NarL-activated narGHJI operon is strongly induced during anaerobic growth in the presence of high concentrations of nitrate, whereas the nrf operon is repressed by nitrate-activated NarL, but induced by nitrite- or nitrate-activated NarP. Synthesis of the cytoplasmic nitrite reductase, NirBD, is induced by both NarP and NarL during anaerobic growth in the presence of nitrate or nitrite. The β-galactosidase assay was used to compare the response of mutants defective in each of these enzymes with the parent strain during growth in the presence of nitrite. Deletion of nrfA or nirB resulted in increased responses to nitrite, suggesting that these nitrite reductases primarily decrease NO accumulation in the cytoplasm, and therefore protect bacteria against nitrosative stress (Table 2). In contrast, the narG mutant responded poorly to the addition of nitrite, consistent with nitrite reduction by NarGHI being the major source of NO in the cytoplasm.

The residual response to nitrite by the NarG mutant indicates that there are additional sources of cytoplasmic NO. To investigate whether this residual induction was due to NO formation by the periplasmic nitrate reductase, NapA, a mutant was constructed that is defective in the nitrate reductases, NarG and NapA. Transcription at Phcp was still induced in this strain during anaerobic growth in the presence of nitrite (Table 2).

Effect of physiologically relevant concentrations of NO on FNR-regulated transcription

Cruz-Ramos et al. (2002) reported that high concentrations of NO derepressed transcription at an FNR-repressed promoter, but repressed transcription at the FNR-activated promoter. It was therefore possible that the lack of derepression of the hcp promoter by externally added NO was due to compensating effects of NO-activated derepression by NsrR and loss of activation by FNR. To determine whether concentrations of NO used in the previous experiments were sufficient to nitrosylate the iron-sulphur centre of FNR and hence, to inactive it, an isogenic pair of fnr+ parental and fnr mutant strains were transformed with two low copy number plasmids from which Phmp::lacZ or a synthetic promoter with a consensus FNR repression site was expressed. Relative to the untreated control, transcription activity at Phmp in the fnr+ strain had increased after 60 min by 24% in response to two additions of 5 μM NO, but there was a slightly greater response of 33% in the fnr mutant (Table 2). The response to NO at Phmp was therefore due to partial relief of NsrR repression rather than relief of FNR repression. Further control experiments with the FNR-repressed but NsrR-independent promoter confirmed that there was no response to NO in either the fnr+ or fnr mutant strains even after further exposure of the cultures to NO, although transcription activity at this promoter was almost fourfold higher in the absence of FNR repression, as expected (Table 2).


The development of a β-galactosidase-based assay to detect NO-dependent relief of NsrR repression has enabled several controversies in the nitrosative stress literature to be clarified. First, although there is a growing consensus that enteric bacteria produce NO mainly as a side product of the reduction of nitrite by NarGHI, some authors have proposed or assumed that NirBD or NrfAB are the major catalysts of NO formation. Data from the transcription response assay are consistent with the membrane-associated nitrate reductase, NarGHI, being the major enzyme involved in the conversion of nitrite to NO. However, nitrite still induced increased Phcp expression even in a narG mutant, suggesting that there must be at least one more protein that catalyses the conversion of nitrite to NO. In contrast to NarG, the periplasmic nitrate reductase, NapA, contributes very little to NO generation. It is possible that this is a side activity of another molybdoprotein. Data in Table 2 also show that Phcp transcription is derepressed more by nitrite in mutants defective in ΔNirBD and NrfAB, presumably because more NO is generated in mutants defective in nitrite reduction to ammonia. This confirms the protective roles of these enzymes against nitrosative stress, but whether they are also minor sources of NO remains to be determined.

An unexpected result was that NO added externally at the highest concentration that did not significantly prevent growth failed to relieve NsrR repression. The 5 μM NO used was considerably higher than concentrations reported to occur physiologically in vivo, which typically are in the nM range and rarely exceed 1 μM (Palmer et al., 1987; Cardinale & Clark, 2005 and references cited therein). A possible exception to this statement is the report that Salmonella within macrophages might be exposed to up to 10 μM NO (Raines et al., 2006). However, nitrite was a more effective inducer of Phcp expression than growth-inhibitory concentrations of 10 or 20 μM NO added repeatedly at 30 min intervals. The smaller and slower response to NO was not due to the rapid decomposition of NO by oxygen because separate experiments with an NO-sensitive electrode confirmed that NO was stable under the anaerobic conditions used. Note that the high pKa value of nitrous acid means that at physiological pH, nitrous acid diffuses across the cytoplasmic membrane, and nitrite can be transported by at least three mechanisms, NarK, NarU and NirC (see, e.g. Jia et al., 2009).

Three of the obvious possible explanations for the minimal response of the hcp promoter to external NO are that derepression of NsrR was counter-balanced by loss of transcription activation by FNR; that derepression of the NsrR regulon resulted in sufficient capacity to repair nitrosative damage to FNR as rapidly as it occurred or that the capacity of the bacteria to reduce NO was sufficient to prevent its cytoplasmic accumulation. Control experiments with the Nsr-independent promoter, FF-37.5::lacZ, eliminated the first possibility and hence, by inference also, the second explanation (Table 2). The results of these experiments also challenged claims that FNR can function as a physiologically relevant sensor of NO (Cruz-Ramos et al., 2002; Corker & Roole, 2003; Pullan et al., 2007).

Although the periplasmic nitrite reductase, NrfAB, was the obvious candidate to provide protection against externally added NO by catalysing its reduction to ammonia in the periplasm (as proposed by van Wonderen et al., 2008), externally added NO still did not induce Phcp::lacZ transcription in a nrfAB deletion mutant as effectively as nitrite. The 10-fold higher rates of NO reduction than nitrite reduction by strains defective in both NirB and NrfA suggest that E. coli has a greater capacity to reduce NO than to produce it from nitrite. We recently reported that even in the absence of all currently characterized NO reductase activities, anaerobic cultures of E. coli still reduce NO rapidly (Vine & Cole, 2011). The data in the current study therefore reinforce our previous conclusion that a significant NO reduction activity remains to be characterized. We favour the explanation that this activity prevents significant damage to cytoplasmic proteins by concentrations of externally generated NO relevant to pathogenicity.


We thank Merve Yasa for help with some of the control experiments.