Field-realistic exposure to neonicotinoid and sulfoximine insecticides impairs visual and olfactory learning and memory in Polistes paper wasps

Insecticides, such as imidacloprid and sulfoxaflor, are widely used chemicals applied to crops through seed, soil and foliage treatments. Once applied, these insecticides are incorporated into plant tissues and become toxic to insects that feed on the plant (Simon-Delso et al., 2015). Neonicotinoid insecticides, such as imidacloprid, are the best-studied and mostly commonly used insecticide type in the world. In the past decade, some neonicotinoids have been banned in outdoor agricultural settings by the European Union because of concerns about detrimental effects in non-target taxa (European Commission, 2013). The sulfoximine-based insecticide sulfoxaflor is a potential replacement for neonicotinoids, and its global use is increasing (Centner et al., 2018). Both insecticides target nicotinic acetylcholine receptors and can impact the sensory integration, learning and memory functions of the mushroom bodies in insect central nervous systems (Bacci et al., 2018; Cabirol and Haase, 2019; Déglise et al., 2002; Martelli et al., 2020; Palmer et al., 2013; Zars, 2000) as well as impairing motion detection (Rigosi and O'Carroll, 2021). Exposure to insecticides can have dramatic sublethal effects on beneficial insects (Siviter and Muth, 2020).

Much research on the sublethal effects of insecticides in non-target organisms has focused on neonicotinoid exposure in bees. Many studies have shown that olfactory learning and memory is impaired in harnessed honeybees and bumblebees exposed to low doses of neonicotinoids (Iqbal et al., 2019; Mustard et al., 2020; Siviter et al., 2019; Tan et al., 2015). Learning is essential to foraging performance, so reduced learning is often detrimental to both colony growth and pollination services (Siviter et al., 2018). Harnessed olfactory learning provides a well-controlled method for assessing learning, but wild bee behavior involves cues in multiple sensory modalities and movement. As a result, there has been increasing focus on insecticide effects on learning in multiple sensory modalities (Ludicke and Nieh, 2020; Muth et al., 2019), natural foraging behavior (Barascou et al., 2022; Muth and Leonard, 2019; Siviter et al., 2021) and motor abilities (Capela et al., 2022).

While low doses of insecticides have detrimental effects on Apis and Bombus bees, less is known about insecticide effects on other non-target insect taxa. A wide range of non-target insects are exposed to insecticides, including non-bee pollinators, predatory insects and detritivores. Given the neurological effects of neonicotinoids and sulfoxaflor (Bacci et al., 2018; Déglise et al., 2002; Palmer et al., 2013; Zars, 2000), we might expect many non-target taxa to exhibit learning and memory deficits. To our knowledge, learning and memory effects of insecticides have not been tested on insects other than bees, but many other sublethal insecticide effects have been identified. For example, monarch butterfly, Danaus plexippus, larvae reared on milkweed sprayed with neonicotinoids have reduced survival (Knight et al., 2021) and cabbage butterfly, Pieris brassicae, larvae exposed to low doses of imidacloprid develop into smaller adults than controls (Whitehorn et al., 2018). Additionally, Nesidiocoris tenuis fed prey treated with sulfoxaflor had decreased offspring production and longevity (Wanumen et al., 2016) and Lasius niger ants fed low doses of imidacloprid were less effective foragers (Thiel and Köhler, 2016).

Wasps provide a good model for investigating the effects of neonicotinoid pesticides on non-target organisms. Aculeate (stinging) wasps are a widespread and diverse group that provide numerous ecosystem services, including pollinating ecologically and economically important plants as well as regulating arthropod populations through predation (Brock et al., 2021). Aculeate wasps also provide a useful comparison to bees because wasps diverged from bees over 100 million years ago (Peters et al., 2017). This study focused on Polistes fuscatus paper wasps, a small colony social wasp found throughout the Eastern United States. Polistes are likely exposed to insecticides though nectar and caterpillar consumption. Like bees, Polistes learn odor and visual cues during foraging (Richter, 2000), so insecticide effects on learning may have detrimental effects on pollination services, predation and colony success.

In this study, we assessed the effects of acute sublethal doses of imidacloprid and sulfoxaflor on P. fuscatus paper wasps. Wasps received a single, oral dose of one of four treatments: control, low-dose imidacloprid (0.5 ng), high-dose imidacloprid (2.0 ng) or sulfoxaflor (20 ng). Then, P. fuscatus were trained to discriminate between two color or two odor stimuli in a free-crawling learning assay. Short-term memory for the colors and odors was assessed by measuring the number of correct choices wasps made 45 min after initial training (Pardo-Sanchez and Tibbetts, 2022; Tibbetts et al., 2019a,b; Weise et al., 2022). We assessed how insecticide exposure influenced wasps' ability to complete the learning and memory assay, the accuracy of a wasp's choices and individual survival.

Polistes fuscatus (Fabricius 1793) foundress wasps and their nests were collected in southeastern MI, USA, and brought back to the lab at the University of Michigan. The wasps were housed with their nest and nestmates in individual boxes and provided with water, sugar and Galleria mellonella larvae ad libitum. Wasps used in this experiment were workers that emerged after the nests were collected. Nests were monitored for workers three times a week and new workers were individually marked using non-toxic Testors enamel modeling paint on their wing tips and returned to their nests. Workers remained on their nests until they were at least 7 days old. Wasps were between 7 and 30 days old at testing.

Imidacloprid and sulfoxaflor were obtained from Sigma-Aldrich and LGC Group, respectively, and dissolved in acetone to make a 1 mg insecticide ml⁻¹ acetone stock solution. Both stock solutions were stored in amber bottles in the refrigerator. Aliquots of the stock solutions were added to a 50% (w/w) sucrose solution to obtain the following treatment group concentrations: imidacloprid low dose (IMDL) 0.5 ng per wasp, imidacloprid high dose (IMDH) 2.0 ng per wasp, sulfoxaflor (SUL), 20 ng per wasp. Reports of field-realistic ranges for imidacloprid in a day's nectar load are 0.024–3.50 ng (Byrne et al., 2014; Cresswell, 2011) or 1–50 ppb (Goulson, 2013) and 1.4–142 ng for sulfoxaflor (Arnet, 2022; Barascou et al., 2022; Capela et al., 2022). These estimates consider measured nectar residue concentrations of the insecticides as well as typical daily nectar and/or pollen consumption. The dosages used in this study are also within the range used in other studies for both imidacloprid (Iqbal et al., 2019; Mustard et al., 2020; Muth and Leonard, 2019; Muth et al., 2019; Siviter et al., 2021; Tan et al., 2015) and sulfoxaflor (Barascou et al., 2022; Capela et al., 2022; Cartereau et al., 2022). For the control treatment, equal aliquots of acetone were added to 50% (w/w) sucrose solution. High- and low-acetone control treatments were used. The low-acetone control matched the volume of acetone aliquot used for the IMD treatments (2 nl of acetone dissolved in 5 µl of sucrose solution per wasp). The high-acetone control matched the volume of acetone aliquot used for the SUL treatment (20 nl of acetone dissolved in 5 µl of sucrose solution per wasp). There was no difference in performance between high and low acetone (high acetone mean=7/10, low acetone mean=6.5/10; t₁₂=0.55, P=0.59). Treatment solutions were stored in amber bottles in the refrigerator. New treatment solutions were prepared every other day and were never used more than 48 h after preparation.

On the day of the learning assay, wasps were isolated from their nests and starved for 1 h before being fed 5 µl of their treatment solution via pipette. Treatment groups were assigned randomly, and feed solutions were administered blind and only revealed after testing had concluded. All wasps included in the study consumed the entire 5 µl dose, resulting in the following treatment group sizes: control: color N=14, odor N=10; IMDL: color N=12, odor N=10; IMDH: color N=12, odor N=10; SUL: color N=12, odor N=10. After feeding, wasps were held in a dark container with water for 1 h prior to the start of the learning assay. Wasps were labeled with neutral IDs to ensure the trainer did not know the wasps' treatment group during training.

Each wasp was trained to discriminate either odors or colors. Previous work has shown that P. fuscatus wasps readily learn to discriminate both color and odor stimuli (Cely Ortiz and Tibbetts, 2021; Tibbetts et al., 2019b; Weise et al., 2022). The color stimuli used were blue and yellow. Colors were printed on photo paper on a Xerox AltaLink C8035 color printer and cut to fit into the walls of the training chambers. Half the wasps were trained to associate blue with safety, while the other half were trained to associate yellow with safety.

Odor stimuli were prepared as in Weise et al. (2022). The odors used during the learning assay were the alkanes dotriacontane and octacosane. These alkanes are non-volatile hydrocarbons that are readily discriminated by wasps and other social insects (Weise et al., 2022). Half the wasps were trained to associate dotriacontane with safety, while the other half were trained to associate octacosane with safety. Each odor stimulus solution was calculated to have the same molarity (0.00888 mmol l⁻¹) with 4.25 mg of dotriacontane and 3.51 mg of octacosane each serially diluted in pentanes to achieve equal molarities. Odors were added to glass chambers by saturating marking tape with the appropriate stimulus solution and leaving the pentanes to evaporate over 24 h so only the odors remained. The odor-saturated tape was then placed on the walls of the glass chambers. New odor stimuli were prepared every other day and no odor stimuli more than 48 h old were used.

Learning and memory was assessed using an established, negative reinforcement assay (Pardo-Sanchez and Tibbetts, 2022; Tibbetts et al., 2019a,b; Weise et al., 2022). Previous work has shown that training and testing is not harmful, as wasps behave normally and survive for months afterward. For example, in previous work, wasps were trained and tested, then released to found nests and raise offspring (Laub, 2023). Wasps in the present study were trained and tested on their ability to discriminate between a stimulus associated with an unpleasant shock (0.4 V) and another stimulus that was associated with no shock. In this experiment, wasps were trained and tested on either the color stimuli or odor stimuli, so each wasp underwent the learning assay once. Performance was tested at a single time point, 45 min after initial training, so it does not explicitly differentiate between how learning and/or memory contribute to performance. Performance at above chance levels indicates that wasps learn the initial association and remember the association for 45 min. Performance at chance levels means that wasps are unable to form associations and/or are unable to remember the association for 45 min. Deficits in learning, memory or both could contribute to poor performance.

In half of the training trials, wasps were placed on an electrified pad inside a 2.5×4×0.7 cm training chamber and exposed to the incorrect stimulus (CS+) while receiving a shock for 2 min. Shocks were provided by conductive foam electrified by two copper wires connected to a Variac transformer. The 0.4 V shock is not harmful to the wasps but is aversive. In the other half of the training trials, wasps were placed on non-electrified foam inside a training chamber of the same size while being exposed to the correct stimulus (CS−) for 2 min. Wasps rested in a dark container with water for 1 min between training trials. The sequence of one CS+ and one CS− training trial was repeated five times for five CS+ and five CS− trials in total. The wasps then rested in a dark container for 45 min before being tested.

After training, we tested how accurately wasps discriminate between the CS+ stimulus and the CS− stimulus during 10 testing trials. Performance was measured as the number of correct choices out of 10 trials. Testing was performed in a 3×10×0.7 cm non-electrified rectangle. One side of the rectangle had the correct stimulus (CS−) and the other side had the incorrect stimulus (CS+). At the start of the testing trial, the wasp was placed in the middle of the rectangle facing neutrally rather than toward one side or the other, the partitions were removed, and the wasp was allowed to walk to either side of the rectangle. Wasps were scored as making a choice when they moved into the 2.25 cm ‘choice region’ on either side (Fig. 1). Wasps were then removed and allowed to rest for 1 min in a dark container. This procedure was repeated 10 times for a total of 10 testing trials. The location of the stimuli was randomly alternated between each trial to ensure the wasps' choices were based on the stimuli and not the location of the stimuli. There was no reinforcement during the 10 trial test, so wasps did not learn during the test alone. As a result, wasps that experienced the 10 trial test without initial training performed at chance level (n=15, mean=4.9, P=0.93). Wasps typically make rapid choices (3 s) (Weise et al., 2022). However, insecticide treatment seemed to interfere with the wasps' ability to walk in the rectangle. As a result, we stopped each trial at 2 min. There were 7 wasps that were unable to complete all 10 testing trials because they stayed in the center of the rectangle and never walked the short distance to either side (5 IMDH, 2 SUL). We analyzed the results twice, once excluding the individuals that did not finish and once including those individuals.

After the training and testing were complete, wasps were placed in individual containers with sugar and water for observation. Unlike many social insects, wasps survive well outside the colony environment (Tibbetts and Banan, 2010). Wasps were checked every day for survival and death dates were recorded for wasps that died in the first week after they were treated with either insecticide or control.

Analyses were performed in SPSS v.28. We compared learning across wasps using a general linear model. The dependent variable was the number of correct choices (out of 10). The independent variables were insecticide treatment (categorical: control, IMDH, IMDL, SUL), type of training (categorical: visual, odor), and the two-way interaction between insecticide treatment and type of training. We used Fisher's least significant differences post hoc pairwise analyses to compare learning between treatment groups. We used binomial tests to assess how performance in each treatment group and stimuli differed from the 50:50 random expectation. The binomial test provides an exact test of whether the number of correct versus incorrect choices differs from the 50:50 random expectation. Binomial tests provide P-values with no test statistics. We performed two different sets of learning analyses. The first included all wasps. The second included only the wasps that were able to complete all 10 trials of the test. Pearson Chi-squared analysis was used to test whether insecticide treatment was linked to whether or not wasps were able to complete all 10 trials of the test. Pearson Chi-squared analysis was also used to test whether insecticide treatment was linked to survival during the first 7 days after treatment.

Treatment with insecticide affected the wasps' performance, as the number of correct choices differed between insecticide treatment groups (Fig. 2; F_3,82=7.7, P<0.001). Post hoc pairwise analyses show that control wasps learned and retained the association better than wasps in the IMDH (P <0.001) and SUL (P=0.001) treatment groups. Wasps in the IMDL group also performed better than individuals in the IMDH (P=0.001) and SUL (P=0.022) treatment groups. However, there was no difference between the control and IMDL groups (P=0.33) or between the IMDH and SUL groups (P=0.33). The type of training (color versus odor) did not influence choice accuracy (F_1,82=0.01, P=0.90). Further, there was no interaction between insecticide treatment and training type (F_3,82=0.58, P=0.63), indicating that there was no difference in the effect of insecticide treatment on color versus odor learning. Binomial tests show that the control wasps performed better than expected by chance for both the color and odor assays (color P=0.0001, odor P=0.0066). For the IMDL treatment group, wasps performed better than expected by chance for the color assay (P=0.022), but not for the odor assay (P=0.088). Wasps in the IMDH and SUL treatment groups did not perform better than expected by chance for either learning assay (IMDH: color P=0.120, odor P=0.057; SUL: color P=0.120, odor P=0.764).

We then tested the factors that influenced the wasps' ability to finish the learning assay. Insecticide treatment influenced the wasps' ability to complete the learning assay (χ²=10.787, P=0.013), with wasps in the IMDH treatment group less likely to complete the learning assay (standardized residual=2.5). Wasps in the control and IMDL treatment group were more likely to complete the learning assay (standardized residual, control=−1.4, IMDL=−1.3). Test type (odor or color) did not affect whether or not wasps completed the assay (χ²=0.775, P=0.379).

In a second set of training analyses, wasps that did not complete all 10 testing trials were excluded from the analysis (n=7, 5 IMDH, 2 SUL). In wasps that completed all the testing trials, treatment with insecticide affected their performance, as the number of correct choices differed between insecticide treatment groups (F_3,75=4.4, P=0.007; mean±s.e.m. control=6.542±0.301, IMDL=6±0.302, IMDH=5±0.437, SUL=5.05±0.380). Post hoc pairwise analyses show that learning and memory was significantly impaired in both the IMDH (P=0.004) and SUL (P=0.003) treatment groups when compared with the control. However, there was no difference in performance between the control and IMDL treatment groups (P=0.26). Further, there was no difference in performance between the insecticide treatments (IMDL versus IMDH P=0.06; IMDL versus SUL P=0.06; IMDH versus SUL P=0.93). Type of training (color versus odor) did not influence choice accuracy (F_1,75=0.19, P=0.68). Further, there was no interaction between insecticide treatment and training type (F_3,75=0.191, P=0.902), indicating that there was no difference in the effect of insecticide treatment on color versus odor learning. Binomial tests show that the control wasps performed better than expected by chance for both the color and odor assays (color P=0.00012, odor P=0.00664). For the IMDL treatment group, wasps performed better than expected by chance for the color assay (P=0.022), but not for the odor assay (P=0.088). Wasps in the IMDH and SUL treatment groups did not perform better than expected by chance for either learning assay (IMDH: color P=0.752, odor P=0.738; SUL: color P=0.920, odor P=0.764).

Insecticide treatment also influenced whether or not wasps survived for 1 week after testing (Fig. 3, χ²=9.871, P=0.020). Wasps treated with SUL were the most likely to die (standardized residual=2.1), while control wasps were the least likely to die (standardized residual=−1.7). Type of training (visual versus odor) was not linked with survival (χ²=0.38, P=0.38).

Our data revealed that a single, field-realistic dose of imidacloprid and sulfoxaflor has detrimental effects in Polistes wasps. Wasps treated with these insecticides performed at chance levels and also made fewer correct choices than control wasps (Fig. 2). Wasp performance was tested at a single time point, so poor performance of sulfoxaflor- and imidacloprid-treated wasps could be caused by the insecticides producing deficits in learning, memory or both. Both visual and olfactory learning and memory were similarly impaired by neonicotinoid treatment, which is notable as some previous work suggests that neonicotinoids may have stronger effects on olfactory learning than visual learning (e.g. Muth et al., 2019). Neonicotinoid treatment also had effects beyond learning. Wasps treated with high-dose imidacloprid were less likely to complete the learning assay than wasps in the other treatment groups. Wasps in the high-dose imidacloprid treatment group remained stationary and unmoving in the testing area, suggesting that high-dose imidacloprid may interfere with motor control. Finally, wasps treated with sulfoxaflor were more likely to die in the week after treatment than wasps in the other treatment groups (Fig. 3). Therefore, treatment with one, field-realistic dose of neonicotinoids has substantial negative effects on Polistes wasps.

Reduced learning and memory in wasps exposed to imidacloprid and sulfoxaflor is broadly consistent with previous work in Apis and Bombus bees (Table S1). A meta-analysis of over 100 studies found that field-realistic doses of insecticides have negative effects on odor learning and memory in harnessed Apis bees, with larger effects on learning and memory at higher doses (Siviter et al., 2018). Insecticides also reduce learning and memory in Bombus bees, with the strength of the effects varying across dosage, insecticide types, and methods used for testing learning (Muth and Leonard, 2019; Muth et al., 2019; Siviter and Muth, 2022). Our results indicate that imidacloprid has dose-dependent effects in Polistes, as learning and memory was not impaired for the wasps in the low-dose imidacloprid treatment group (0.5 ng per wasp) but were impaired for the wasps in the high-dose imidacloprid treatment group (2.0 ng per wasp). Our results suggest that insecticide effects on learning may be widespread across insect taxa rather than being confined to bees.

In contrast with some previous work on bees, we found that insecticide treatment had similar effects on wasp visual and olfactory learning (Fig. 2). Neonicotinoid impact on bumblebee learning is modality specific, with stronger detrimental effects on olfactory learning than on visual learning (Muth and Leonard, 2019; Muth et al., 2019). Modality-specific effects may occur because of how neonicotinoids target bee neural systems. Neonicotinoids act on nicotinic acetylcholine receptors expressed throughout the insect brain (Grünewald and Siefer, 2019), including mushroom bodies, an area of the brain involved in learning, memory and sensory integration (Bacci et al., 2018; Déglise et al., 2002; Palmer et al., 2013; Zars, 2000). There are multiple subtypes of nicotinic acetylcholine receptors that may be differentially expressed in the olfactory and visual system, leading to differential insecticide effects on visual versus olfactory learning (Moffat et al., 2016). Polistes wasps diverged from bees over 100 million years ago, so wasps may have different receptor distributions in the mushroom bodies (Jones and Sattelle, 2010), leading to differences in how insecticides affect visual and olfactory learning in bees and wasps. Notably, although we found no significant evidence of a modality-specific effect on performance, there was a hint of a modality-specific difference in the IMDL treatment group. Wasps treated with IMDL learned the color stimuli but not the odor stimuli. Additional research will be useful to assess whether there could be a dose-dependent effect of imidacloprid on different learning modalities.

Imidacloprid also had detrimental effects on the movement of treated wasps. Wasps treated with high-dose imidacloprid were less likely to finish the testing assay than wasps in the other treatment groups. Instead, imidacloprid-treated wasps remained stationary in the testing chamber until they were removed after 2 min. Remaining stationary is extremely unusual during testing. Wasps typically move rapidly in the testing chamber, making choices in an average of 3.15 s (Weise et al., 2022). This unusual wasp behavior is consistent with evidence in other taxa that imidacloprid causes impaired motor function and coordination. High doses of neonicotinoids cause severe motor difficulties such as trembling, uncoordinated movement and hyperactivity (Matsuda et al., 2020). Field-realistic doses have more subtle, but important effects. For example, bumblebees were less motivated to forage and experienced bouts of hyperactivity followed by reduced or no movement after imidacloprid exposure (Muth and Leonard, 2019). Bees treated with imidacloprid were also slower to complete their learning task than control bees (Mustard et al., 2020). Additionally locusts treated with imidacloprid had impaired collision avoidance via decreased jumping escape (Parkinson et al., 2020) and/or flight escape behavior (Parkinson et al., 2017). Effective foraging requires a high level of motor coordination, so even a small reduction in motor function can have dramatic effects on individual and colony success (Barascou et al., 2022; Feltham et al., 2014; Siviter et al., 2021; Williamson et al., 2014).

Surprisingly, our study also found that wasps treated with sulfoxaflor had lower survival than wasps from other treatment groups. The median lethal dose (LD₅₀) of sulfoxaflor is estimated to be between 96 ng per bee (Cartereau et al., 2022) and 146 ng per bee (ESFA, 2014) in the honeybee Apis mellifera. The dose used in this study (20 ng per wasp) was 14–20% of the A. mellifera LD₅₀. As A. mellifera and P. fuscatus have similar body sizes, it is initially surprising that relatively low doses of sulfoxaflor caused increased wasp mortality. Similar doses did not increase mortality in A. mellifera (Barascou et al., 2022; Capela et al., 2022; Cartereau et al., 2022). One possibility is that training may be somewhat stressful, such that the combination of sulfoxaflor treatment and training could increase wasp death rates beyond what we would see with sulfoxaflor treatment alone. Training and testing alone had no effect on wasp survival, as no control wasps died in the week following training. Further, previous work has shown that trained wasps find nests, raise offspring and survive for months following training without any reduction in survival (Laub, 2023). Nevertheless, the combination of training and insecticide treatment could contribute to reduced survival. Sulfoxaflor in combination with other stressors has been shown to increase bee mortality, while exposure to sulfoxaflor alone does not (Siviter et al., 2020). Pollinators often face multiple stressors in the wild that may greatly increase the costs of insecticide exposure, e.g. heat stress, nutritional stress, parasites, multiple insecticides (Botías et al., 2021; Goulson et al., 2015). Additional work on the mortality costs of sulfoxaflor across contexts will be useful to understand the real mortality burden of insecticides in the wild.

This study differs from previous work on sublethal insecticide effects by focusing on wasps. Most current work examines honeybees and bumblebees, but there is a growing appreciation that studying insecticide effects on diverse taxa is essential (Braak et al., 2018; Franklin and Raine, 2019). The limited information available indicates that pesticide effects are variable, with some taxa being more sensitive to insecticide than others (Franklin and Raine, 2019). In addition, highly social bees such as Apis may be more buffered against negative insecticide effects than taxa with small colonies, such as Polistes wasps, or solitary species (Sgolastra et al., 2018). In addition to providing an interesting comparison to bees, wasps are an economically important group in their own right. Wasps are extremely diverse and may surpass bees in species richness and abundance. Many wasps act as pollinators and non-bee pollinators are essential, as they account for 39% of visits to crop flowers (Rader et al., 2016). Wasps also hunt a range of arthropods and play an important role as bio-control agents (Southon et al., 2019). Wasps may be exposed to insecticides through eating insect prey and the combination of exposure via nectar and predation may result in relatively high insecticide exposures for wasps and their larvae (Wumuerhan et al., 2020).

Overall, understanding the effects of insecticides in diverse taxa is crucial because arthropod decline may cause broad ecosystem challenges because of the crucial role of arthropods in the food web (Wagner, 2020). Our study provides further evidence that sublethal exposure to neonicotinoid insecticides can have significant negative effects on non-target species. Future work in additional taxa and contexts will be important to understand the breadth of insecticide effects on both diverse insects and ecosystem function.

Thanks to Micah Golan and Anna Vi for help with wasp collection and care. Dr Emily Laub and Juanita Pardo Sanchez provided feedback on earlier versions of the manuscript.