When trying to contemplate natural products research, the first images that may come to mind are possibly from the media: Sean Connery striding across his hut in the Amazonian rainforest to check the latest LC/MS data (a la the film Medicine Man) or video of a snorkeler hunting a tropical reef for a poisonous snail (from a source like National Geographic perhaps).
Wade through the pages of the Journal of Natural Products and you can be forgiven for assuming that efforts to convert environmental metabolites into medicines have never been stronger.
Yet, for many large pharmaceutical companies, the revolutionary discoveries like Selman Waksman and streptomycin are seen as moments of past glory. They are events to be commemorated, as with the 2015 Nobel Prize for Medicine for the discovery of the anti-malarials avermectins and artemisinin as wonderful serendipities.
“The great discoveries of the past were made come-by-chance or by happenstance,” concurs Nathan Magarvey, McMaster University researcher and Adapsyn Bioscience founder. “They didn't occur in any kind of directedness.”
“There's a beauty in that, there's a serendipity in it,” he continues. “But it does not lend itself to a clean-headed think about what to do and when and how.”
“There are many, many success stories, but given the risk involved in getting a drug lead through this very lengthy, very expensive process, Big Pharma is shying away from engaging in the early steps of drug discovery,” he adds.
For Andy Haigh, Adapsyn CEO, it comes down to the notion of productivity.
“You can get a hit,” he explains. “Then you have to deconvolute it. Spend months working on things only to say, hey, I've discovered this thing and it's been seen before or it's not that interesting or the chemistry is not that good or it's not drug-like.”
Despite the diversity of secondary metabolites that could be probed, he continues, only a small proportion have the potential to be developed as drugs.
“And at the same time, you saw combinatorial chemistry and DNA-encoded libraries [DELs] and all these other things, which were that much more predictable or easier to work with,” Haigh recalls.
By the 1990s, industry had grown weary of rediscovering the same molecules over and over, whether that was due to the metabolite being the most highly expressed or its bioactivity overwhelming signals from less active constituents.
Where was the chemical diversity that everyone had been promised?
“Major pharmaceutical companies diminished or even terminated their [natural products] programs and shifted their focus to target-based drug discovery, which relies on HTS [high-throughput screening] of synthetic chemical libraries,” explained Wayne State University’s Ping Dou and colleagues in a 2019 editorial. “Unlike phenotype screens, target-based approaches are based on the validated understanding of the interaction of tested compounds with the designated targets and thus are considered to have higher hitting rates ... Unfortunately, this strategic shift correlated with the overall reduction in novel lead compounds in the drug development pipeline and the substantial decline in new drug approval.”
Not everyone has given up on the natural product space, however. New technologies and approaches are allowing researchers to tap deeper into the elusive chemical diversity, offering a glimmer of a new generation of transformative discovery.
“We're not trying to reinvent the modality of natural product, but we're trying to really improve the discovery process by looking at capturing all the biosynthetic diversity out of the environment,” says Brad Hover, vice president of Discovery Research and Platform Technology at Lodo Therapeutics. “And then also improve the production process by looking at things such as the tractability for both discovery and the scale-up of those molecules.”
Old school, new effort
Rather than reinvent the harvest-extract-analyze wheel, some researchers like Barry O’Keefe and colleagues at the National Cancer Institute (NCI) and Leidos Biomedical Research are focused on finding ways to improve the wheel.
As they explained, the NCI Program for Natural Product Discovery (NPNPD) has a mandate to facilitate sample prefractionation and develop integrated resources for rapid isolation and structure identification of biologically active natural products. In early 2020, they described an automated fractionation process using solid-phase extraction followed by HPLC.
“In an HTS context, this workflow is able to process 500 primary hits and provide in excess of 10,000 wells containing dried-down pure and semi-purified compounds back to the screening lab in a two-week window at a quantity that yields a sufficient amount of subfractions for multiple secondary assays,” the authors highlighted.
The researchers then analyzed active wells with a combination of NMR, LC/MS, and FTIR, comparing the spectroscopic and spectrometric data against in-house and commercial databases to identify the biologically active components and permit dereplication of known compounds.
“Using this approach on all of the hits, we found that most subfractions contained pure or semi-purified compounds in quantities that enabled collection of high-quality analytical data and subsequent rapid dereplication of known compounds,” the researchers noted. “In total, we dereplicated the structures of 28 known biologically active natural products as well as identified subfractions that contained new natural products.”
They were also able to identify and isolate moderately active compounds that might have been missed in crude extracts or whose activity might be masked by more active components.
In other ways, natural products research is evolving with the convergence of newer technologies that are allowing researchers to explore new questions.
“The metabolomics field has really made a very important contribution to the natural products drug discovery process,” says Kerr.
After putting fermentation extracts through an initial biological screen, whether for desired activity or off-target toxicity, he says, his lab routinely performs a metabolomic screen to determine which of the extracts contain molecules with known activity. The goal is to fail those samples rapidly.
“In my academic lab or any early screening lab, we can run this LC/MS, dump the data into a database and filter out the stuff that we believe has known chemistry, and thereby focus efforts on the fractions that appear to have new chemistry,” he explains.
And combining uHPLC with high-resolution mass spectrometry (HRMS) has allowed this work to proceed on an unprecedented scale. Where a typical HPLC run might take 30 to 45 minutes, he says, uHPLC has shorted that cycle time to five or six minutes.
“If you’re thinking one or two samples, that's not a big deal,” Kerr presses. “But when a lab is screening hundreds or thousands of extracts, you can multiply 1,000 times five minutes, and that's manageable. When we’ve got 1,000 samples times 45 minutes, you can't do it.”
This throughput is essential for Kerr as he explores samples from marine environments.
“The marine environment is really exciting to anybody interested in natural products for two reasons,” he enthuses. “One, there's much greater biodiversity in our oceans than on land, whether we're looking at the macro-organisms—sponges, tunicates, corals—or most of us now are working with microbes, perhaps from these invertebrates.”
Secondly, the oceans have been less explored than land habitats.
“The vast majority of the marine natural products work has come from zero to 30 meters,” he adds. “I've been on some submersible dives going into 2,000 feet, but that's very expensive, it's slow, and it's not particularly productive.”
Expanding the biodiversity into which we tap is critical, as the greater the biodiversity in one’s collection of genetic resources, the greater the chances of finding new chemistry.
And greater exposure to these resources is changing how we understand them, Kerr continues, pointing to sponges as an example.
“Sponges have traditionally been the most prolific resource for natural product discovery,” he explains. “What we've learned in the last 10 to 20 years, however, is that the sponges aren't the chemists.
“Sponges are these really unique habitats for very complex microbial assemblages. And it's these microbes within sponges—same goes for corals, tunicates, and other marine invertebrates—that are the real chemists.”
But just as metabolomics methodologies are improving because of advances in uHPLC and HRMS, microbe-based natural products research can still be hamstrung by an old nemesis: culturing the unculturable.
Kerr and others describe this challenge as the great plate count anomaly.
“Look under the microscope,” he says. “You can see thousands of microbes. But when you try to culture them, you can culture maybe 1 percent of what's there.”
Recognizing the importance of the environmental factors to microbial growth, Kerr and colleagues developed the ichip or isolation chip, whereby a solution of bacteria isolated from homogenized sponge, for example, is suspended in agarose and applied to a small plastic plate in which 100 holes serve as isolation chambers.
“The idea was there should be on average one bacterium per hole in that plate,” Kerr explains. “And on each side of that plate, we put membranes of porosity such that bacteria can’t travel back and forth, but the molecules can.”
The plates are then incubated in the sponges to provide the bacteria with the chemical signals that the researchers do not yet comprehend and cannot replicate in the lab.
“The hope is that for bacteria that wouldn't grow in a Petri dish, that single cell in the little hole in the plastic plate will become 105, 106,” he continues, adding that these larger colonies can then be harvested as starter cultures back in the lab.
In 2019, Kerr and his colleagues described one such effort that led to the discovery of several new bacterial species, including one that produced a novel tyrosine derivative with gram-positive antibacterial activity.
One of the drawbacks to the ichip, however, is that although the microbes can sense their chemical environments, they cannot sense each other.
To address this, the researchers developed the microbial domestication (MD) pod, whereby single bacteria are encapsulated in agarose microbeads and then cultured together in their native environment. The goal was to recapitulate the entirety of the natural signaling experiences.
“It's a great collaboration between my own lab and Ali Ahmadi, who's a bioengineer at the University of PEI,” Kerr remarks. “He has 3D printing and microfluidics technology to be able to establish little miniature beads with one bacterial cell, put them back in a device containing 1,000 or so beads, and then put that back into the environment.”
Kerr, Ahmadi and colleagues described the initial design and characterization of MD pods this past November, enthusiastically anticipating but not yet demonstrating novel chemistries.
Beyond simply improving microbial growth, however, such bacterial co-cultures may also become important to exploring the full biochemical repertoire of these organisms, as many microbial biosynthetic gene clusters (BGCs) remain silent until activated.
“A Streptomyces will have the genetic ability to biosynthesize, say, 25 natural products,” Kerr explains, “but when we use our standard fermentation conditions, that organism will produce two or three.”
He continues that 85 percent of our antibiotics and many other drugs originated from bacterial or fungal sources and yet this might only represent 10 percent of available species. What is possible if that number could be doubled or tripled?
“Any bacterium living in any habitat is not living in axenic culture,” Kerr suggests.
Rather, it is living among thousands of other microbes, and in the case of marine microbes, may be exposed to host factors produced by the sponge or coral.
“We don't understand these environments very well, and that's likely in large part responsible for our inability to turn on these silent or cryptic gene clusters,” he continues.
For this reason, Kerr and many others have turned to microbial co-culture where a “producer” bacterium is challenged by an “inducer” bacterium. As in the wild, it is hoped the producer could then generate chemistries to ward off or respond to the inducer.
In metabolomics studies, Kerr’s lab is incrementally examining microbial pairs to see what pathways are activated.
Here again, however, culturing can be a challenge as it can be difficult to identify growth conditions that are optimal for both paired microbes. To address this, Kerr has used heat-killed inducers such that researchers only need to optimize growth conditions for the producer strain.
Last June, they compared the two strategies for cryptic BGC expression, finding that both methods demonstrated good reproducibility and produced different metabolic profiles from each other.
“The ability of each co-culture method to aid in the discovery of new natural products was validated by isolation and structural characterization of two new natural products, N-carbamoyl-2-hydroxy-3-methoxybenzamide and carbazoquinocin G,” the authors wrote, further suggesting that the heat-killing and more traditional co-culture approaches should be viewed as complementary.
Before coming to Canada, Kerr founded a center of excellence in marine biotechnology while at Florida Atlantic University. This work eventually led to the founding of Nautilus Biosciences, which has also since moved to the University of PEI campus.
The company’s initial focus centered on pseudopterocins, molecules derived from bacteria found in octocorals, which have shown significant anti-inflammatory properties. In vitro, these molecules have shown efficacy versus both triple-negative breast cancer and pancreatic ductal carcinoma, acting as a glucocorticoid receptor-alpha agonist to inhibit tumor cell proliferation and invasion.
More recently, the company and collaborators described the isolation and identification of the novel polyketide/non-ribosomal peptide levesquamide from a Streptomyces strain. The compound showed strong antimicrobial activity against Mycobacterium tuberculosis, including drug-resistant strains, but showed no cytotoxicity against Vero cells, suggesting its therapeutic potential.
“Nautilus was set up in PEI with the idea of moving away from focusing on one family of natural products to try to build a very exciting library or collection of bacteria from rarely studied habitats such as octocorals,” Kerr recalls.
The activity led to collaboration with and ultimately acquisition by Croda, a multinational chemical company better known in pharma for excipients rather than active pharmaceutical ingredients.
“Working with them, our library grew from maybe a few hundred interesting bacteria to several thousand bacteria and fungi,” Kerr adds. “So, we now have a very substantial, very well characterized, novel marine microbial library.”
Croda has also invested heavily in the screening infrastructure at Nautilus, accelerating and deepening their discovery efforts. But just as importantly, Kerr stresses, is Croda’s sense of corporate social responsibility.
“I just set up revenue sharing agreements with governments in developing countries and parts of Canada,” he says, offering the example of an agreement with Nunavut Tunngavik.
“Croda absolutely insists that there is such an agreement in place so that when something is commercialized for one of the Nautilus microbes, a small amount of money goes back to the people of Nunavut,” he shares.
Whereas Kerr has augmented “old school” natural products research with newer technologies, other groups are moving microbial fermentation further down the development pipeline. Instead, they look to leverage other omics approaches and artificial intelligence to optimize novel chemistry discovery.
For companies like Lodo Therapeutics and Adapsyn Bioscience, the first steps in natural product research aren’t culturing microbes and identifying biological activity but rather sequencing microbes and identifying BGCs.
“The issue here is, how do we reinvent the process of discovering those molecules?” explains Lodo CEO Dale Pfost. “We do so with the advent of the tools of the trade: high-throughput sequencing, molecular biology breakthroughs, synthetic biology.”
Based on work from Rockefeller University scientist and company co-founder Sean Brady, Lodo’s platform uses metagenomics and bioinformatics to screen the DNA of environmental microbes for interesting and novel BGCs, based on homology to existing clusters and their natural products. The approach allows them to build a phylogenetic tree of related clusters in other organisms, offering a diversity of analogues to the original product.
In a 2016 review, Brady, Micah Katz, and Brad Hover offered two ways in which this phylogenetic organization could provide insights.
“Congeners of characterized natural products can be mined by pursuing gene clusters associated with NPSTs [natural product sequence tags] that group closely with the biosynthetic domains encoding for known natural products,” the authors wrote. “Conversely, potentially novel natural products can be pursued by focusing on gene clusters associated with NPSTs that do not group closely with any previously characterized sequences in these families.”
As Pfost explains, however, this effort to identify analogue diversity may prompt questions of how this is different from combinatorial chemistry.
“Maybe it's analogous,” he responds, “but here's the big difference: That combinatorial chemistry library didn't have any assurances of biological relevance.”
“In this case, when you have a family of related close cousin bacteria that are producing slightly different molecules, 100 percent of them are biologically relevant,” he continues. “Those bugs aren't making those molecules for nothing.”
Lodo’s chief scientific officer and senior vice president, Steve Colletti, takes the sentiment one step further, suggesting that the chemical structures produced by nature are inconceivable to medicinal chemists.
“These synthetic libraries essentially try to model or copy those attributes of natural product chemical space, and it comes up short,” he says. “All you hear about is these big numbers, and in those big numbers, what you get is big redundancy.”
Reduced costs of sequencing and increased throughput have been vital to the development of robust predictive tools, says Hover.
“Some of our generalized machine learning tools, the kind that predict chemotypes, are only enabled recently because of the advancements in the sequencing, because we have more data to train on,” he explains. “And that's been huge for us.”
Because many of these BGCs run 40 to 140 kb in length, advances in long-read sequencing have been vital to the Lodo process, adds Colletti, pointing specifically to Oxford Nanopore technology.
BGC size has also proven challenging when the company tries to move the BGCs to other bacterial hosts for heterologous gene expression.
To clone the BGC, Lodo and several other groups have developed methods to initially isolate smaller overlapping BGC fragments that they reassemble in yeast to form bacterial artificial chromosomes (BACs). Once reassembled, the BGC-encoding BACs are then transferred into host bacteria, optimized for BAC replication, heterologous expression and culture growth.
Simply transferring a BGC from one organism to another may not be sufficient for expression, however. As noted earlier, BGCs can be silent or cryptic, and depending on the source, may have significantly different regulatory machinery.
“Now that we have more understanding of the gene cluster and the various different regulators in there, we sort of understand more about what promoters need to get dropped in to switch on a silent gene cluster,” Hover responds. “You can do it even in a native system, as well, where you go through and figure how to turn on these silencing clusters.”
In 2018, Hover and colleagues at Rockefeller University and Rutgers University applied this process to the isolation and heterologous expression of soil bacteria BGCs for a distinct class of antibiotics called malacidins.
The 10-membered cyclic lipopeptides showed antibacterial activity against MRSA as well as other multidrug-resistant microbes and sterilized MRSA-infected wounds in a rat model. Also, unlike many other antibiotics, the malacidins did not appear to select for resistance in vitro.
More recently, Northwestern University’s Neil Kelleher and colleagues used a similar approach but with fungi, developing fungal artificial chromosomes to identify, clone, and express a BGC from Aspergillus terreus that generated terreazepine. Although they didn’t describe the biological activity of this secondary metabolite, the authors highlighted novelty in its biosynthesis and saw this as a proof of concept for new areas of natural product exploration.
Being able to readily access these novel BGCs and metabolites doesn’t mean, however, that Lodo is going to create a vast library of compounds.
“We're trying to focus it around the predictions and the targets that we want to pursue,” says Hover.
Key to this predictive capacity will be the further expansion of the company’s in-silico activities. This push was the impetus behind Lodo’s recent acquisition of Conifer Point.
Conifer, and its predecessor BioLeap, specialize in using cheminformatics and structural biology to develop molecular models and automating the docking of thousands of compounds in silico, identifying optimal binding partners.
“The addition of Conifer Point’s technology allows de-novo discovery starting from a disease target of interest, rather than a known drug scaffold,” Pfost explained in announcing the acquisition. “This makes it possible for Lodo to discover, enrich, and prioritize large numbers of biologically and pathway-relevant molecules addressing hard-to-drug disease targets in silico, with unprecedented gains in predictive power and significantly enhanced efficiency.”
Even more recently, Lodo acquired Hibiskus BioPharma and signed licensing agreements with UC, Riverside and Michigan State University for the preclinical proteasome inhibitor TIR-199 (now LODO-141), a member of the syrbactin natural product family. Aside from pursuing the lead itself, which has shown efficacy against a variety of cancer cell lines as well as hollow fiber and mouse xenograft models, Lodo will also search for analogues using its BGC methodology.
“We also have another internal project in the Wnt pathway that essentially started in Lodo,” adds Colletti, highlighting a second program versus solid tumors.
He also notes partnerships with Genentech to develop gram-negative antibacterials and with the Gates Foundation targeting tuberculosis.
Also working heavily on the predictions side of natural products and BGCs is Adapsyn, built largely on the work of Nathan Magarvey.
“I was very interested in how to embrace natural products in a modern way,” Magarvey recalls. “Started off thinking about it as a microbial genomics exercise, which is what the field has mostly tagged along with, and then worked with [Harvard University’s] Chris Walsh, who's sort of the godfather of natural product biosynthesis and the enzymology of this.”
As his work continued, Magarvey realized that it wasn’t going to be enough to understand the genome side of things, but that you would also need to be able to predict the molecules that resulted from these biosynthetic pathways.
“[That] prediction process is not an easy thing because you have to catalogue all the enzymes, you have to relate those enzymes into reactions, you have to manifest those reactions in some process,” he explains. “And you have to relate those predicted products to all the things that have been found in the past.”
That early work garnered attention from a Big Pharma player, leading to Adapsyn’s launch in 2016.
“The company was originally set up based on inbound interests from Pfizer in work that was coming out of Nathan's lab,” explains Andy Haigh, company CEO. “They had a strain collection, which is now at Scripps, and the idea fundamentally was to profile a number of those strains and figure out if those strains are indeed producing novel chemistry.”
When starting with genomes, as opposed to extracts, the company focuses on unusual or novel BGCs.
“We generally try to work with bacterial strains that have more than one novel gene cluster, because if we're going to go through all the effort of fermenting them and pulling the compounds, we'd rather get more than one per unit effort,” Haigh continues.
Equally important, though, is the effort to generate structure predictions based on those genomes, giving the company a sense of the “drug-likeness” of the resulting metabolite(s).
These efforts are exemplified by two recently described analytical platforms that combine bioinformatics and machine learning with sequencing and biophysical characterization.
As Magarvey and colleagues explained in 2019, DeepRiPP is an automated platform to discover novel ribosomally synthesized and post-translationally modified peptides (RiPPs), which offer significant chemical diversity but present unique analytical challenges.
Unlike the BGCs described earlier, the pathway to produce RiPPs does not always involve chromosomally adjacent genes. Thus, the researchers developed a two-stage deep-learning module that first searched genomes for possible RiPP precursor peptides and then predicted their likely cleavage patterns. They then used a local alignment algorithm to dereplicate the potential RiPPs, prioritizing only those that suggested novel chemistries.
The final step involved parsing metabolomic data from crude extracts to match genomically encoded RiPPs with MS peaks for subsequent purification and structure elucidation. When validated against 20 bacterial strains predicted to contain RiPP precursors, DeepRiPP was able to identify three novel molecules, two of which were purified and shown to have structures identical to those predicted by the platform.
PRISM 4, meanwhile, predicts both structure and biological activity of secondary metabolites based solely on microbial genome sequences.
“Although it is now straightforward to identify clusters of genes responsible for secondary metabolite biosynthesis, translating between genome sequence and the complete chemical structures of the natural antibiotics encoded therein represents a key challenge, and one that has taken on an increasing importance in an era of growing global antibiotic resistance,” the authors wrote. “PRISM 4 represents the most comprehensive effort to address this challenge to date.”
Against a set of 1,281 known BGCs, PRISM 4 was able to identify 1,230 (96 percent) and generated at least one chemical structure for 1,157 (94 percent) of those BGCs. Furthermore, in a collection of 3,759 dereplicated complete bacterial genomes, PRISM 4 identified 22,446 BGCs and at least one predicted structure for 7,404, including “metabolites such as β-lactams, alkaloids, phosphonates, cyclodipeptides, bisindoles, and aminoglycosides.”
The researchers then tested whether the platform could help evaluate what specific biological activity the predicted metabolites might have. To do this, they curated bioactivity data from 1,281 BGCs in the literature and trained support vector machines to predict the probabilities that any given BGC product would have antibacterial, antifungal, antiviral, antitumor, or immunomodulatory activity.
Using this trained quantitative predicted structure-activity relationship (QPSAR) model on a set of 10,000 complete or metagenome-assembled genomes, “PRISM 4 identified 1,589 BGCs producing antibacterial compounds, 331 antiviral BGCs, 289 immunomodulatory BGCs, 272 antifungal BGCs, and 248 antitumor BGCs, in addition to a further 1,055 BGCs with more than one predicted biological activity.”
Given its design, the researchers were quick to recognize two key limitations of PRISM 4: it cannot identify BGCs from undescribed families nor products of novel enzymatic activities, and its utility is currently limited to prokaryotic organisms.
“We can do machine learning to predict what structural patterns mean, in terms of activity,” says Magarvey. “So, you can build this whole compendium of information that really further sharpens the pencil of the researcher toward getting at the compounds that will matter. At least the new ones, quickly.”
Thus, rather than clone clusters, Adapsyn has the multi-omics that allow the interrogation of organisms as they are, as they make the natural product.
This latter point is important to Haigh.
“I know that there are groups out there that make the argument that you need to do heterologous gene expression to access silent gene clusters and all of that stuff,” he says. “We don't necessarily subscribe to that notion because we see a lot of novel chemistry in the native producers.”
“Heterologous gene expression is potentially a really powerful technology, and there's certainly a place for it,” he continues. “But to a large extent, you end up potentially swapping out one inefficient process for another inefficient process.”
Instead, Adapsyn ferments the native producers at milliliter-scale, giving them enough starting material to confirm the presence of the predicted metabolite and to start the biological assays.
“Because we know where it is, because we have the size, we can isolate it,” Haigh explains. “We can put it into a plate, we can get it relatively pure early on, and then we can start to do biological activity assays that ultimately guide what we do downstream.”
“We've done some pretty comprehensive phenotypic screening that allows us to start bucketing compounds based on toxicity, based on potential indication, and we can do all of that stuff before we've actually scaled up and solved the structure,” he continues.
The ability to screen early in the process has dramatically improved the team’s productivity, he says, allowing them to access and screen more than 1,000 novel compounds in a year.
Only once they are confident they have something interesting does the team scale up to larger cultures to give them enough material to solve the metabolite’s structure using LC/MS and NMR, as well as continue to test it using orthogonal assays.
“I will say, when we call something novel up front and then go solve it downstream, we’re correct 96, 97 percent of the time,” Haigh offers, adding, “We can typically go from a genomic or metabolomic analysis and prioritization on the front end to a solved structure in six weeks, and most of the time in there is fermentation or chemistry time.”
Magarvey is excited about the opportunities ahead.
“We're breaking free from the normal organisms that we've said are important,” he enthuses. “We're looking at everything with these tools and predicting the molecules from all those different sources. Even with those organisms we studied well, we can put the novel compounds in wells.”
From Haigh’s perspective, there is an irony that new technologies are not so much reinventing the wheel as reimagining it.
“Weirdly enough, we are using really modern technology to facilitate a really old version of drug discovery,” he says. “You're talking about a 50-year-old process of taking a bug, growing it out, doing bioactivity-based fractionation and all that stuff. We are essentially revisiting that but using very modern computational tools to essentially make it as productive as it used to be in the past.
“That's I think where we want to go, and I think collectively where the field wants to go."