.*Karageorgi, M., *S.C. Groen, K.I. Verster, J.M. Aguilar, F. Sumbul, A.P. Hastings, J.N. Pelaez, S.L. Bernstein,
T. Matsunaga, M. Astourian, G. Guerra, F. Rico, S. Dobler, A.A. Agrawal & N.K. Whiteman (2019).
Genome editing retraces the evolution of toxin resistance in the monarch butterfly. Nature.
Link to Viewable PDF and Downloadable PDF *equal contribution.
Inspiration for the Monarch Fly
To illustrate the framework for our overall study, Dr. Marianthi (Marianna) Karageorgi was inspired by Salvador Dalí's whimsical Rosa Papilio and created this composition that includes wings from the milkweed butterfly lineage, and an outgroup, attached to the stem of milkweed (Apocynaceae: Asclepias curassavica), which hosts the caterpillars of the top three butterflies. The butterfly wings from the base of the stem to the flower belong to the orange-spotted tiger (Mechanitis polymnia), the striped blue crow (Euploea mulciber), the queen (Danaus gilippus) and the monarch (Danaus plexippus), respectively. The butterfly evolved sets of amino acid-changing substitutions in the gene encoding the sodium pump that provide increasing levels of resistance to toxic cardiac glycosides present in milkweeds.
To the upper left of the flower, a CRISPR-Cas9-engineered Drosophila melanogaster "monarch fly" carries the same set of substitutions at three amino acid residues in its sodium pump as the monarch butterfly and is as resistant to cardiac glycosides as the monarch. M. polymnia does not feed on Apocynaceae plants, but rather on species of Solanaceae. Nonetheless, its genome encodes L111 in the first extracellular loop of the sodium pump, which, as we discuss below, confers some resistance to cardiac glycosides.
Concept and Art Direction by Dr. Marianthi Karageorgi and Dr. Noah Whiteman, Art Production by Eugene Randolph Young, Photographs by Dr. Peter Oboyski/Essig Museum and Dr. Marianthi Karageorgi.
To illustrate the framework for our overall study, Dr. Marianthi (Marianna) Karageorgi was inspired by Salvador Dalí's whimsical Rosa Papilio and created this composition that includes wings from the milkweed butterfly lineage, and an outgroup, attached to the stem of milkweed (Apocynaceae: Asclepias curassavica), which hosts the caterpillars of the top three butterflies. The butterfly wings from the base of the stem to the flower belong to the orange-spotted tiger (Mechanitis polymnia), the striped blue crow (Euploea mulciber), the queen (Danaus gilippus) and the monarch (Danaus plexippus), respectively. The butterfly evolved sets of amino acid-changing substitutions in the gene encoding the sodium pump that provide increasing levels of resistance to toxic cardiac glycosides present in milkweeds.
To the upper left of the flower, a CRISPR-Cas9-engineered Drosophila melanogaster "monarch fly" carries the same set of substitutions at three amino acid residues in its sodium pump as the monarch butterfly and is as resistant to cardiac glycosides as the monarch. M. polymnia does not feed on Apocynaceae plants, but rather on species of Solanaceae. Nonetheless, its genome encodes L111 in the first extracellular loop of the sodium pump, which, as we discuss below, confers some resistance to cardiac glycosides.
Concept and Art Direction by Dr. Marianthi Karageorgi and Dr. Noah Whiteman, Art Production by Eugene Randolph Young, Photographs by Dr. Peter Oboyski/Essig Museum and Dr. Marianthi Karageorgi.
|
The Herb of Death (and Life)
Dame Agatha Christie was a pharmacist in WWI and knew about medicinal plants, such as foxglove (Digitalis purpurpea). She used foxglove as the murder weapon in her novella, The Herb of Death, and cleverly weaved the fact that foxglove is a biennial into the plot. The plant's rosette leaves look like leaves of sage to the untrained eye. Who done it? The gardener? The cook? The server? The host? On a lighter note, William Withering determined that foxglove could be used directly to treat 'dropsy' or symptoms of congestive heart failure. In fact, the cardenolide (a type of cardiac glycoside) toxins digitoxin, digoxin and ouabain are still used to treat heart conditions. Cardenolides bind to the 1st extracellular loop of the animal sodium-potassium ATPase, a critical ion pump that maintains the resting membrane potential of cells. This pump uses up a large fraction of the ATP molecules in our brains by pumping sodium out (using ATP) and potassium into the cell. The amino acid sequence of this pump is widely conserved in animals and therefore a good target for a plant toxin. The toxin likely evolved to deter herbivores. Plants from several different orders, including species that are very common house and garden plants like tropical milkweed (Apocynaceae), pussy-ears (Crassulaceae) and lily-of-the-valley (Asparagaceae) synthesize these compounds, as do toads (e.g., Bufo which make compounds called bufadienolides) that store them in large glands behind their eyes. They are also made by fireflies (discovered by Tom Eisner here) that are brightly colored during the day and bioluminescent at night. Fireflies of the femme fatale species even obtain their own toxins by eating the males of species that produce them and are attracted to their flashes. It is possible that fireflies indeed obtain cardiac glycosides from milkweeds but this is still an open question (see here). |
Bufo photo from Creative Commons (https://commons.wikimedia.org/wiki/File:Bufo_bufo_on_grass2.JPG)
|
|
Toxic host plants and insects Monarch butterfly (D. plexippus) caterpillars feed exclusively on the leaves, stems, and flowers of milkweed plants. This one, to the right, is resting in my garden, in plain sight, on a milkweed species native to Africa. If you look closely, you can see smaller, orange-colored insects on the stems, called oleander aphids (Aphis nerii). These aphids drink the toxic sap of the same plants. Why are the monarchs and the aphids so brightly colored? They are advertising their toxicity to predators. Such so-called 'warning coloration' or aposematism is found in many other animals that are toxic (e.g., skunks, poison frogs) or venomous (e.g., coral snakes, scorpionfish, wasps). One can easily imagine this evolving to the benefit of both prey and predator: toxic or venomous animals that didn't evolve it are killed at a higher frequency than those that had some coloration to warn would-be predators. Vertebrate predators (birds at least) are by and large innately avoidant of brightly colored objects and those that get a taste of a monarch wing learn to never do so again as you might remember from the 'barfing bluejay' story whereby a naive blue jay eats a monarch and then promptly vomits (based on numerous studies by the late Professor Lincoln Brower and colleagues, including Professor Linda Fink). The bird vomits due to the cardiac glycosides triggering an emetic response. Think ipecac syrup...
A side story is why the larva and adult (below, right) are brightly colored (aposematic) but the chrysalis (middle right) is not, but is "bright green with a golden diadem" to quote Miriam Rothschild. Of course the chrysalis is toxic, but monarch caterpillars, when they are about to form a chrysalis and begin metamorphosis, leave their milkweed host to find a secluded place, sometimes a large leaf or a stem under which they attach their cremaster. Why isn't the chrysalis brightly colored? The likely adaptive explanation is that while the larva and adult have means of escaping would-be predators who might want to nip at them, the pupae are immobile and as a result, highly vulnerable to any mechanical perturbation. While a larva can fall off of a leaf and an adult can still fly well with a bit of wing missing, the chrysalis is extremely fragile. Thus, life history evolution worked to build in a cryptic 'middle' of the insect's life cycle, despite the toxicity. Note that during the day or so before the adult emerges, the chrysalis does turn black as the adult colors take hold, betraying its presence. You can read about the experiments that led to this insight here. The distasteful monarch is largely free of many of the threats unprotected butterflies face from avian predators as they forage, mate and oviposit. There can be little doubt that the monarch's toxicity is associated with its great migration from the prairies of eastern North America to the Oyamel fir forests of Mexico each year--a 5000-mile trip in some cases . Although birds and mice do feast on monarchs as they lose their toxicity over time in these overwintering grounds, yet another natural experiment that reveals how the toxins protect them until they begin to dissipate. Resistance to the toxins sequestered by the monarch are the pillars supporting the subsidiary adaptations we all know and love, from the bright orange and black and white polka dotted bodies, to their remarkable migratory abilities, that we are just beginning to understand. Who discovered exactly what makes monarchs distasteful? Another Dame at the top of her game: none other than Miriam Rothschild... |
Photos copyright Noah Whiteman (top two) and Shane Downing (bottom)
|
|
Sequestration Dame Miriam Rothschild and three colleagues showed in 1968 (here) that monarch butterflies did indeed sequester heart poisons from milkweeds they consumed as larvae. Dame Rothschild was one of the preeminent entomologists and parasitologists of her time and one of the few women elected to the Royal Society. She never formally attended school, discovered how fleas synchronize their life cycles with rabbits by relying on rabbit hormones after taking a blood meal, discovered how fleas jump, worked to decriminalize homosexuality in Britain, invented automobile seat belts and thought of the monarch butterfly as a natural wonder of the world. There are three must-watch videos of her here. The monarch appears on the scene around 7:20 in the first, and then continues in videos 2 and 3. I highly recommend watching all three videos to get a sense of her brilliant mind. |
|
Toxin Resistance Mechanism?
In 1992, a paper was published here by Ferdinand Holzinger by Ferdinand Holzinger and colleagues who measured sodium pump activity from the monarch as well as the tobacco hornworm (a control species not adapted to cardiac glycosides) across increasing doses of a cardiac glycoside called ouabain (highly water soluble). They found that the monarch's sodium pump activity was not altered by the presence of ouabain, but the control species’ pump’s activities were. When they obtained a sequence of the amino acids in the 1st extracellular loop, where ouabain was thought to bind to the pump, they found that a histidine was present at position 122, instead of an asparagine in non-adapted species. They suspected that this change might alter the binding affinity of ouabain. But it took time before the tools were available to begin to test this hypothesis in cell lines and then later, in whole animals. Holzinger and Fink in 1996, proved this in vitro here. Then, a remarkable discovery in convergence appeared in the literature in 2012.
In 1992, a paper was published here by Ferdinand Holzinger by Ferdinand Holzinger and colleagues who measured sodium pump activity from the monarch as well as the tobacco hornworm (a control species not adapted to cardiac glycosides) across increasing doses of a cardiac glycoside called ouabain (highly water soluble). They found that the monarch's sodium pump activity was not altered by the presence of ouabain, but the control species’ pump’s activities were. When they obtained a sequence of the amino acids in the 1st extracellular loop, where ouabain was thought to bind to the pump, they found that a histidine was present at position 122, instead of an asparagine in non-adapted species. They suspected that this change might alter the binding affinity of ouabain. But it took time before the tools were available to begin to test this hypothesis in cell lines and then later, in whole animals. Holzinger and Fink in 1996, proved this in vitro here. Then, a remarkable discovery in convergence appeared in the literature in 2012.
|
Parallel Discovery of Parallel Evolution in Milkweed Specialists
In 2012, a landmark paper was published in PNAS here by Professor Susanne Dobler, Professor Anurag Agrawal and colleagues. They found that when they aligned the sodium pump amino acid sequences from many species of insects that independently evolved to be specialists on cardiac glycoside-producing plants, parallel amino acid changes were apparent at several sites, including 111 and 122, in the 1st extracellular loop. Often a leucine, valine or threonine evolved at position 111 and histidine frequently re-evolved at position 122, as in the monarch, across many of the cardiac glycoside specialists. A few months after that paper was published, my friend Professor Kailen Mooney and I published a News & Views in Nature here. We pondered that it was remarkable how a few mutations might have such reverberating consequences in the ecological communities that they evolved in resulting in cascading effects of resistance mutations to keystone toxins (a term coined here). We also proposed that in order to determine if these mutations were truly sufficient for cardiac glycoside resistance, they would need to be tested in whole animals (in vivo) and the vinegar fly (Drosophila melanogaster) was a great candidate. A whole-animal approach is important to understand the costs as well as the benefits of the mutations. Costs at the organismal level may not be apparent at the level of the cell or enzyme. Older (but sophisticated) tools to enable allele swapping (so called knock-ins) existed at the time and CRISPR-Cas9 was also just coming online as yet a new way to conduct such studies. It turns out we were a bit naive when suggesting this, given how difficult it turned out to be. Just after the the Dobler et al. and Whiteman and Mooney News and Views were published, another elegant study (below right), conducted by Dr. Ying Zhen, Professor Molly Schumer, Professor Peter Andolfatto and colleagues at Princeton University, was published in Science here. Zhen et al. found patterns very similar to the ones reported by the Dobler et al. study with the important addition that the genomes of some of the cardiac glycoside specialists contained more than one copy of the gene encoding the sodium pump (paralogous copies). This suggested either subfunctionalization or neofunctionalization was at play. What was more, some of these gene copies encoded more of the putative resistance substitutions in the sodium pump, and others had fewer (were more similar to the ancestral sequence). They found that in some species, each paralog was expressed in a tissue-specific manner, with the more resistant copies being more highly expressed in tissues exposed to the toxins and the more 'susceptible' copies more highly expressed in tissues protected from the toxins. This was later recapitulated in other species, adding a layer of complexity that continues to generate new insight into the molecular basis of adaptation. |
|
|
The Path to Monarch Flies In 2012, Professor Agrawal had read our News & Views and invited me (to the right) to Cornell University to give a talk at a co-evolutionary biology symposium. He asked if I wanted to work with him to 'do' the gene editing experiments for gain of function in D. melanogaster with him and Professor Dobler. They could provide expertise in the ecology and biochemistry if we could do the genetics. I had just started my lab at the University of Arizona and wasn't sure if starting a side project like this, as interesting as I thought it would be, was a good idea. Nonetheless, it was the height of the Great Recession, and I didn't have a major grant yet. So, we applied and received funding for three years in 2013 to push the project. Our goal was to focus on the amino acid substitutions that evolved along the evolution of the milkweed butterflies, which had been studied in great detail at that point by Professor Agrawal and his students, including (now) Professor Georg Petschenka. They showed that resistance had evolved step-wise in the nymphalid butterflies that fed on milkweeds, culminating in the monarch, but with transitional character states present in extant species like the common crow butterfly and the queen butterfly. They published this work here and again, a similar result was published even earlier by Peter Andolfatto's group here, who reported the first evidence for convergent resistance to cardiac glycosides in the sodium pump at the 111 and 122 positions between the monarch and a beetle. I then made one of the best decisions of my career, and hired Dr. Simon "Niels" Groen (a former mentee of mine when I was a postdoc at Harvard) as a postdoc to help lead the genome editing part of the project. Dr. Groen arrived in Tucson and began designing the RNA guides and developing a strategy with the fly embryo injection service we contracted with (GenetiVision in Houston, TX--I have no financial or other interest in the company). Our goal was to simply attempt to knock-in the milkweed butterfly substitutions at 111 and 122, and use PCR-RFLP to screen for the mutations. It sounded so easy, but it didn't work, and we didn't know why. Were the mutations lethal in the fly? Were there off-target mutations? Was it just inefficient? Were we just unlucky? We troubleshot for a while and depleted most of our grant funds. Then, we met up with the other PIs at Cornell when Professor Dobler was doing a sabbatical there for a last-ditch discussion, and decided to introduce another mutation, at position 119 (from A119 to S119). This mutation is associated with changes at position 122 in cardiac glycoside specialists, but didn't stand out because it is often found outside of species that feed on these plants (illuminated by Ph.D. candidate Julianne Pelaez). We thought it might play a role in mitigating the apparent cost of changes at 122. At the same time, we decided with the injection service that a PCR-based screen was not practical. But I was moving to Berkeley in 2015 just as our funding was running out, and wanted to let go of the project because I just didn't think I could keep investing in it. Dr. Groen, however, did not want to give up and asked if we could try "one more time" although he too had moved onto another postdoc position. I had, fortuitously, in 2016 just received generous new funding (an R35 grant for early stage investigators) from the NIGMS of the NIH. Without Dr. Groen's faith and that funding, this project would have ended. But, we soldiered on and opted for a two-step design in which we first used CRISPR-Cas9 editing to knock a GFP cassette (that made the flies’ eyes glow green) into the region of the sodium pump gene we wanted to edit. Once we were able to do this, our 'deletion' line enabled us to screen for loss of green eyes after a second step of CRISPR in which our donor plasmid with the various butterfly mutations was swapped with the GFP cassette. In what seemed like a blink of an eye, our strategy began to work. Fortunately, a new and outstanding postdoctoral fellow, Dr. Marianthi (Marianna) Karageorgi, who got CRISPR-Cas9 editing up and running in D. suzukii here had just arrived in my lab and wanted to help finish the project with Dr. Groen (who is doing a postdoc with Professor Purugannan at NYU). But "finish" meant *starting* the actual phenotyping, which involved organizing and implementing the actual experiments (involving tens of thousands of flies) to test the benefits and costs of the mutations and then making even more new mutants. Dr. Karageorgi also led, with undergraduate Michael Astourian and Dr. Geno Guerra, the Reproducibility of Mutational Order analysis above (third from top, below the phylogeny). Dr. Karageorgi suggested we collaborate with Dr. Fidan Sumbul and Professor Felix Rico to conduct molecular docking experiments to begin to understand the biochemical basis of the toxin resistance mutations. The final homozygous viable mutants were *very* difficult to obtain. Almost all of our second-round positive flies would make it to the very last cross, but once made homozygous, the flies would then die. Fortunately, with brute force, we eventually obtained viable mutants for all of the transitional character states that evolved in the milkweed butterfly lineage at positions 111, 119 and 122, and eventually two single mutants (A119S and N122H) that didn't evolve in the lineage alone, but were critical for understanding the ordered path of mutations. The most insightful experiment tracked survival of the immature stages to adulthood--from the first through third larval instars and pupal stage, to adult eclosion (figure to the immediate right). This is because it is the monarch caterpillars, not the adults, that feed on plants producing the heart poisons. However, adult butterflies sequester as well, and so our experiments on adult survival were also critical (figure to the right). Early on in the adaptive walk, some toxin resistance is provided by L111, but a lot more is provided through positive beneficial epistasis for resistance by the combination of L111 and S119 (the 'crow fly’; next figure down). These mutations together result in a fitness "ratchet" (for larvae), that likely pulled early milkweed caterpillars up the flank of the fitness peak, far enough to colonize milkweeds (this is speculative). The next mutation to evolve, L111V on top of the A119S already there (the 'queen fly') provided the same level of resistance in whole animals, but a slightly (though not significantly) higher level of resistance in their pumps. The 'monarch fly' with VSH at 111, 119 and 122 was not killed (and in fact was not affected negatively at all) with 30 mM ouabain, an otherwise lethal dose. Thus, VSH was sufficient for full resistance (the summit of this fitness peak) to the toxin at two levels of biological organization: immature development and adult survivorship assays as well as at the physiological level in pump assays using neural tissues (conducted by Amy Hastings and Professor Agrawal). The VSH flies were as resistant to ouabain as actual monarch butterfly pumps (see figure below). Sufficiency of monarch-level resistance for the the VSH mutant was then independently confirmed in Sf9 insect cells by Professor Dobler. When we examined just the S119 mutation alone, some resistance was provided in larval-adult and adult survival assays. When H122 was examined in a single mutant, it provided as much resistance as the VSH genotype in larval-adult and adult survival assays. However, in physiological assays, it provided less than that found in the VSH flies, which was also consistent with the molecular docking scores obtained by Dr. Sumbul and Professor Rico. These physiological and in silico experiments on the biochemical mechanism helped illuminate why and how the VSH triple mutant took the crown (sorry, couldn't help it) in the evolution of the monarch butterfly! Although I didn't think this next experiment would work given all of the other secondary compounds in milkweeds (i.e., it would be a bridge too far), Dr. Karageorgi put dried milkweed leaves into fly food and found that the VSH flies emerged at much higher rates than the control flies. This means that the VSH mutations may actually enable the big ecological leap to feeding on milkweed plants. It was a surprising result (and an example of why it is important for mentees to push back when their mentor is skeptical) of an idea. Finally, Amy and Professor Agrawal measured ouabain levels in flies and the VSH flies retained trace amounts of ouabain after emergence from their puparia (which is the same life stage as the chrysalis of butterflies), suggesting that the VSH mutations when in larvae unlock a route to sequestration of the toxins. One might think this is trivial, but it is quite possible that all ouabain could have been lost during larval molts or larval-puparium transition (toxins are lost in these processes in actual monarch butterflies, see the study by Frick and Wink here). The resistance mutations may unlock the door to passive sequestration and facilitate the evolution of full resistance to the levels sequestered by the monarch. When the L111 and H122 mutations were on their own, each exhibited severe neurological problems (seizures after shaking--see video below right). Remarkably, when L111 was paired with S119 in a double mutant, and H122 with S119 in a double mutant, the cost of these resistance mutations was ameliorated in part by the S119 mutation. The pumps seemed to work just fine physiologically based on tests without ouabain, so without the whole animals we would not have been able to discover the costs. Interestingly, the H122 mutant was by far the most costly, and the most resistant. Our paper reporting the results of all of these experiments focused on the full adaptive walk in the monarch lineage and on its natural history was accepted in principle at Nature on August 5, 2019 (which happened to be my 43rd birthday!) after spending nearly one year in peer review (it was first submitted for consideration to Nature in October 2018). It is the first time (to our knowledge) that CRISPR-Cas9 genome editing has been used to engineer up to three coding mutations in an essential gene in an animal. Our final paper can be found here. Another parallel emerged alongside this story of repeated evolution. At the end of August 2019, Professor Andolfatto's group at Columbia University and collaborators at other universities published a paper in eLIFE here that used other DNA editing methods to study some of the same (and other) single and double mutations in vivo in fruit flies at the three sites and found similar benefits against ouabain and costs for adult fly survival and in physiological preparation of neural tissue assays. In their flies, L111 and H122 were recessive lethal mutations and, like many of ours, led to bang sensitivity phenotypes. They found the costly effects of those resistance mutations were rescued to some extent by S119, which was also found to confer resistance to ouabain in adults, but not in the physiological assays (as in our study). The two studies reciprocally illuminated the nature of the ordered path of these mutations in distantly related insect lineages. Another really nice finding of their study was that they were able to ascertain that the resistance mutations resulted in cardiac glycoside resistance when in the heterozygous state. This helps us understand how the process of adaptation unfolds because natural selection probably first tested the mutations in heterozygotes. If the resistance phenotypes of the mutations were all recessive, it would make it more difficult for them to increase in frequency initially. Lastly, our team was diverse in many ways, it included researchers from the U.S., Japan, Greece, Turkey, Germany, France, and the Netherlands. Eight of the fifteen co-authors are not men. The team included postdocs, undergrads, technicians, and professors. The U.S. team included people from many different places and identities, including immigrants, children of immigrants, grand-children of immigrants, great grand-children of immigrants, and African-American, Creole, Latinx, LGBTQA+ and Asian-American scientists. This resulted in a more innovative and multifaceted study than if we'd had a more homogenous team, in my view (a nice summary is here). Finally, without the leadership of postdocs Dr. Karageorgi and Dr. Groen (the two co-first authors), the project would never have taken flight, or have been finished. Summary: We used convergent evolution as a tool to identify substitutions at three amino acid positions that evolved in concert and in a particular order, during evolutionary transitions to milkweed specialization and toxin sequestration in insects that feed on cardiac glycoside-producing plants. With these candidates in hand, we then retraced a multi-step adaptive walk from beginning to end in the Danainae butterflies using CRISPR-Cas9 to knock them into the fruit fly's native sodium pump gene, and then tested their adaptive value and costs--a first in whole animals. This helped reveal how the monarch butterfly evolved to be fully resistant to the heart poisons in its larval diet, and then to be toxic as an adult, with reverberating consequences for ecology and evolution. Positive intramolecular epistasis for resistance evolved early in the adaptive walk, pushing early milkweed butterflies up the lower flank of the fitness peak. Then, triple mutant "monarch flies" were sufficient for resistance at the whole animal, physiological and biochemical levels to the highest levels of heart poison, allowing flies to complete development when reared on fly food containing ground milkweed leaves. The most resistant single mutation evolved last in the adaptive walk, a likely consequence of costs, including a seizure-inducing one. This constrained mutational path opened up a new niche for an iconic butterfly lineage and underpins their brazen inter-generational migratory behavior from Michigan to Mexico where they stay largely (although not entirely) protected from predators en route, while overwintering and during their return. Many questions remain, so please stay tuned! DISCLAIMER: These are my informal notes on our process of discovery, and may contain unintentional errors. For the full, peer-reviewed paper, please go to the Nature website (https://www.nature.com/articles/s41586-019-1610-8) Written and revised by Noah Whiteman, October and November 2019 |
We created “monarch flies” using CRISPR-Cas9 genome editing to carry the same three mutations in the sodium potassium pump as the monarch butterfly. The sodium pump is the target of cardiac glycoside toxins produced by milkweeds. A mutant "monarch fly stands on a wing of an aposematic monarch butterfly wing. Photograph by Ph.D. candidate Julianne Pelaez (copyright Julianne Pelaez).
A "monarch fly" clings to the wing of a monarch butterfly. The fly carries the same mutations as those found in the monarch butterfly. Photo by Julianne Pelaez.
|