Application of simultaneous selective pressures slows adaptation

Abstract Beneficial mutations that arise in an evolving asexual population may compete or interact in ways that alter the overall rate of adaptation through mechanisms such as clonal or functional interference. The application of multiple selective pressures simultaneously may allow for a greater number of adaptive mutations, increasing the opportunities for competition between selectively advantageous alterations, and thereby reducing the rate of adaptation. We evolved a strain of Saccharomyces cerevisiae that could not produce its own histidine or uracil for ~500 generations under one or three selective pressures: limitation of the concentration of glucose, histidine, and/or uracil in the media. The rate of adaptation was obtained by measuring evolved relative fitness using competition assays. Populations evolved under a single selective pressure showed a statistically significant increase in fitness on those pressures relative to the ancestral strain, but the populations evolved on all three pressures did not show a statistically significant increase in fitness over the ancestral strain on any single pressure. Simultaneously limiting three essential nutrients for a population of S. cerevisiae effectively slows the rate of evolution on any one of the three selective pressures applied, relative to the single selective pressure cases. We identify possible mechanisms for fitness changes seen between populations evolved on one or three limiting nutrient pressures by high‐throughput sequencing. Adding multiple selective pressures to evolving disease like cancer and infectious diseases could reduce the rate of adaptation and thereby may slow disease progression, prolong drug efficacy and prevent deaths.


| INTRODUC TI ON
Methods for slowing the rate of adaptation of cell populations would be valuable tools for preventing human suffering from evolving diseases including both infectious diseases and cancer. The evolution of drug resistance is a major problem worldwide for cancer therapy, antibacterial agents, antifungal agents, antimalarial agents, as well as pesticides and herbicides (Barzman et al., 2015;Hughes & Andersson, 2015;Tabashnik, 1989). Methods for slowing the evolution of drug resistance would thus save lives as well as produce significant economic benefits in agriculture, through improved pest and weed management. There are several known evolutionary mechanisms that may affect rates of adaptation. These include variations in mutation rate (Arjan et al. 1999;Sprouffske, Aguilar-Rodríguez, Sniegowski, & Wagner, 2018), as well as the other parameters of the rate of evolution such as population size and generation time (Maley, Szabo, & Reid, 2011), Hill-Robertson interference affecting evolution at linked sites (Hill & Robertson, 1966;McVean & Charlesworth, 2000), fluctuating selection (Cvijović et al., 2015), epistasis (Chou, Chiu, Delaney, Segrè, & Marx, 2011;Perfeito, Sousa, Bataillon, & Gordo, 2014), and genetic and functional interference, which predict that competition between different clones may alter the rate of adaptation. Functional interference is essentially antagonistic pleiotropy. This occurs when mutations that confer an advantage in one environment have a fitness cost in another environment (Bell 2008). Antagonistic pleiotropy is often invoked as reason for the limits of selection and adaptation (Williams 1957;Orr 2000;Hong and Nielsen, 2013) particularly in explaining the evolution of specialists vs. generalists (Elena, Agudelo-Romero, & Lalić, 2009;Nikolin et al. 2012). Clonal interference theory predicts that multiple beneficial mutations may arise and compete within a population, effectively slowing the time to fixation of adaptive mutations in asexual populations de Visser & Rozen, 2006) such as neoplastic cells. Although a prediction of interference theory, this has rarely been directly investigated. Here, we aim to directly test one possible mechanism for reducing the rate of adaptation: evolution under multiple, competing selective pressures.
Multiple competing selective pressures are undoubtedly of both ecological relevance, as organisms are likely facing multiple pressures at any given time, as well as applicable to understanding both progression and therapy in human disease, including cancer and infectious diseases (Strelkowa & Lässig, 2012). In particular, a cancer accumulates large numbers of mutations and progresses through a complex evolutionary process of clonal selection and expansion (Attolini & Michor, 2009;Greaves & Maley, 2012;Maley et al., 2017;Merlo et al., 2006), meeting many of the conditions required for clonal and functional interference (Illingworth & Mustonen, 2011;Sprouffske et al., 2012). If we can effectively alter the rate of evolution through the application of multiple selective pressures, we can slow tumor progression and the evolution of therapeutic resistance.
Existing empirical studies of clonal interference are of replicate populations in a single environment (de Visser & Rozen, 2006;Gerrish & Lenski, 1998;Kao & Sherlock, 2008;Rozen et al., 2002) and do not directly address the role of simultaneous selective pressures on the rate of evolution. Additionally, evolution in complex environments has infrequently been directly examined experimentally (Barrett et al., 2005;Lunzer et al., 2002).
We have chosen to study this phenomenon in a S. cerevisiae model system, a common system for the development of anti-cancer agents (Pereira et al., 2012;Simon & Bedalov, 2004). Here, our selective pressures are various nutrient limitations: glucose, a carbon source; histidine, an essential amino acid; and uracil, a nucleotide.
These nutrients were chosen because the biosynthesis pathways of histidine and uracil have little overlap and are generally separate from the central metabolic pathways of glucose metabolism.
We have examined the relative rate of adaptation of S. cerevisiae to limiting glucose, histidine, uracil, and the combination of all three during a ~500 generation batch culture evolution experiment. We measure the rate of adaptation by quantifying the increase in fitness over that time period. We sought to reduce the likelihood of "generalist" mutations that confer an advantage under all three conditions by choosing pressures with no known overlap in the biosynthetic pathways of these nutrients. Rather, "specialist" mutations conferring an advantage to each nutrient limitation should be more likely in this experiment. In accordance with the theoretical predictions of reduced rates of adaptation under multiple selective pressures, we observed that populations evolved under multiple pressures were less adapted to growth under any single pressure than the populations evolved under only a single selective pressure.
When a population evolved under a single selective condition, it evolved a higher fitness on that condition than the ancestral strain. This is in keeping with well-established principles of natural mechanisms for fitness changes seen between populations evolved on one or three limiting nutrient pressures by high-throughput sequencing. Adding multiple selective pressures to evolving disease like cancer and infectious diseases could reduce the rate of adaptation and thereby may slow disease progression, prolong drug efficacy and prevent deaths.

K E Y W O R D S
adaptation, cancer therapy, clonal interference, functional interference, slowing evolution selection and has been amply demonstrated by experimental evolution. However, populations evolved under multiple pressures did not statistically significantly improve in fitness relative to the ancestral strain for any of the given pressures. To show that fitness improvements were not due to general adaptation to batch culture conditions, we verified that evolved populations had the highest fitness when grown on the media used for their evolution and in most cases did not show consistent fitness gains in other media. This work provides a direct empirical demonstration that evolution under multiple selective pressures slows the rate of adaptation, a phenomenon critical to our understanding of disease evolution, resistance, and development of novel therapies.

| Long-term evolution
A single colony of an S288c haploid strain of Saccharomyces cerevisiae (MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0) was picked from a plate of YPD medium (20 g/L glucose, 20 g/L peptone, 10 g/L yeast extract, 15 g/L Bacto agar) and grown overnight in SD media (6.7 g/L Bacto Yeast Nitrogen Base, 20 g/L D-glucose) supplemented with 20 mg/L L-histidine, 100 mg/L L-leucine, 30 mg/L L-lysine, and 20 mg/L uracil. Following this, a small inoculum was transferred to 10 ml fresh media with limiting glucose (47.5 mg/L), limiting histidine (0.38 mg/L), limiting uracil (0.325 mg/L), or a combination of all three limitations at the concentrations above.
These concentrations of nutrients were chosen after careful study to determine conditions that limited the growth of the cells to an approximately equal low growth rate and carrying capacity. Each day, 10 ml cultures were diluted 1/8 such that cultures underwent three doublings/day. Frozen stocks were generated at regular intervals by freezing in 25%-33% glycerol at −80°C. We started a second set of limiting cultures (replicate B) using frozen stocks from the first day's growth of replicate A on limiting media, such that the first three generations of evolution occurred in common before the replicates were generated. We would have liked to run more replicates, but due to the labor-intensive and long-term nature of the experiment, requiring daily management without break for 5 months, that was not feasible. We evolved the yeast for a total of 552 generations for replicate A and 507 generations for replicate B.

| Measuring fitness
Batch culture competitions were performed to determine fitness because the carrying capacity of the nutrient-limited cultures is too low to allow for reliable OD readings and calculation of growth rates for fitness measurements. Evolved populations and a G418-sulfateresistant reference strain (MATα his3::kanmx leu2Δ0 lys2Δ0 ura3Δ0) were grown overnight in SD media supplemented with Ura, His, Leu, and Lys as described above. The reference strain was used because there are no differential selectable markers on the evolved populations to allow differentiation between populations in direct competition. Approximately equal numbers of the evolved population and reference strain were mixed, washed, and transferred to fresh media with limiting nutrients. Aliquots of this initial mixture were plated onto YPD plates and incubated 48 hours at 30°C. Batch culture competitions were then incubated for 24 hours at 30°C, 220rpm in a shaking incubator. After 24 hours, the overnight cultures were diluted and plated onto YPD as described. Three plates containing between ~ 15 and 150 colonies each from both 0 and 24-hour time points were replica-plated onto YPD/G418 sulfate and again onto YPD as a transfer control. Plates were incubated overnight and the proportion of evolved population and reference strain colonies was calculated from the total number of reference strain colonies growing on the three G418 sulfate replica plates. Relative fitness (W evol/ ref ) was calculated by taking the ratio of the Malthusian parameters (m) of the evolved population (m evol ) and reference strains (m ref ) (Lenski et al., 1991): where N 0 = initial population size and N t = population size at t = 1 day (24 hours) as determined from colony counts based on dilution series.

| Statistical analysis
Statistical comparisons were performed using R statistics software. We took a conservative approach and used nonparametric Wilcoxon rank sum tests to compare the fitness values with a simple Bonferroni correction for multiple testing, α < 0.05/29 = 0.0017.
Initial tests determined whether A and B replicates were statistically significantly different. Subsequent tests combined replicates where there was no significant difference between the replicates. Illumina GAII reads were analyzed for point mutations and insertions/deletions per methods previously described (Araya et al., 2010).

| High-throughput sequencing
Briefly, an average of 7.49 × 10 6 reads were collected from each of the 9 populations. Each read was aligned individually with Bowtie 0.12.2 (Langmead et al., 2009) (Langmead et al., 2009). The data were smoothed by removing segments less than three standard deviations apart, and only breakpoints containing more than 1,000bps were considered for further analysis. On average, 93.35% of the genome was covered at an average depth of 24.4 reads per base (see Table S1 for sequencing results for individual populations).

| Fitness results
Fitness of all eight evolved populations and the ancestral strain was evaluated on a medium containing either limiting glucose, limiting histidine, or limiting uracil media as well as a medium limiting all three nutrients. All fitnesses were measured relative to a common Ura evol populations on the histidine-limited medium, generated negative fitness values (Figure 1b, d). This implies that cells were not just unable to grow during the 24-hour competition, but that cells actually died during this time period. Trypan blue staining of Glu evol , His evol , and Ura evol on limiting histidine medium shows that, indeed, the cells under conditions that garnered negative fitnesses are inviable (blue) after 24 hours, while those with positive fitnesses are viable, healthy cells after this same time period and staining procedure. Comparisons of scaled relative fitnesses across all media types (Figure 2) show that fitness is highest for each evolved population on its own evolved media (e.g., GluA evol and GluB evol have the highest fitness on limiting glucose) and that the GluB evol population seems to have an unexpectedly high fitness on alternative nutrient limitations. As a control, fitness was also measured on YPD medium (Figure 3). Interestingly, most evolved populations have a lower fitness than the ancestral strain on rich media. Also as a control, the fitness of the ancestral strain was measured on each of the media types (Figure 1a-d).
To analyze the rate of adaptation, we measured the relative fitness of populations evolved under single and multiple selective pressures. Each of the single pressure evolved populations (Glu evol , His evol , Ura evol ) has higher fitness than the ancestral strain on the medium in which they evolved (Table 1). However, the populations evolved under all three pressures did not evolve a significantly increased fitness relative to the ancestral strain under any of the single pressures, with the exception of AllB evol , which appears to have adapted to glucose limitation (Table 1). When compared to each other, populations evolved under single selective pressures also have higher fitness than populations evolved under all three selective pressures (Table 2), with the exception of growth of the Glu evol and AllB evol populations on glucose-limited media, which were not significantly different (Figure 1a). Our experimental observation is consistent with the predictions of the reduced rate adaptation under multiple selective pressures.
Next, we tested whether populations evolved on a particular limitation had higher fitness on that limitation than on other limiting nutrients (e.g., Glu evol populations should have higher fitness than populations evolved on other limiting nutrients when grown on limiting glucose, Table 3). Furthermore, if there are no generalist mutations that increase fitness to multiple limiting nutrients or to batch culture conditions, then a population evolved under one limiting nutrient should have no better fitness than the ancestral strain when grown on a different limiting nutrient (Table 4). We confirmed this prediction in all cases except that populations evolved under limiting glucose evolved high fitness on limiting uracil and vice versa, suggesting that there may be generalist mutations that improve fitness with respect to both uracil and glucose limitation (Tables 3 and 4).

| Illumina high-throughput sequencing results
We sequenced the genomes of the eight evolved populations plus the ancestral strain. There were 675 evolved mutations across the  Table S5), striking evidence of convergent evolution. Eight of 17 mutations in the All evol populations are nonsense mutations. We note that the strain used in our study had an intact genetic pathway for growth on glucose metabolism, but contained the deletions of HIS3 and URA3 genes, which could provide a stringent limit on the ability of these populations to grow without histidine and uracil in the medium; however, we did not see a reduction in the number of mutations on limiting histidine or uracil F I G U R E 1 Relative fitness of eight evolved populations and ancestral strain on single limiting nutrients as well as the combination of three limiting nutrients. Each dot presents an independent fitness measurement from each evolved population. Note the difference in scale (y-axis) in each of the four graphs relative to glucose. We found only one point mutation that went to fixation in the GluA evol population, though the region surrounding a glucose transport gene, HXT6, was amplified. Amplification of HXT6 and another, nearly identical transporter, HXT7 (Kruckeberg, 1996), has been reported under nutrient-limiting conditions (Brown et al., 1998). In GluB evol , a negative regulator of the glucose sensing signal transduction pathway (YDR277C) was truncated via a nonsense mutation. UraA evol and UraB evol populations produce phenotypically distinct colony types, but have the same two underlying point mutations. This likely occurred in the growth of the original clone or in the first day's growth under uracil limitation, before replicates A and B were split. HisB evol has a mutation in HIS7, an essential component of the histidine biosynthesis pathway. This is a surprise given that the gene in the histidine biosynthesis pathway, HIS3, is already deleted in the ancestral strain. However, because HIS7 is immediately upstream of HIS3, loss of function mutation in HIS7 could lower the level of 5'-phosphoribosyl-4-carboxamide-5-aminoimidazole (AICAR), one of the His7p enzymatic products which is expected to accumulate in his3Δ populations and is known to be toxic at high levels (Rébora et al., 2005).

| D ISCUSS I ON
We have shown that we can reduce the overall rate of adaptation by simultaneously applying multiple selective pressures (Tables 1 and 2). The mechanism is consistent with the phenomenon of clonal interference, whereby beneficial mutations arise and compete within an evolving population (Gerrish & Lenski, 1998;Sniegowski & Gerrish, 2010); this phenomenon has also been examined in sexual populations (Felsenstein, 1974;Hill & Robertson, 1966;Roze & Barton, 2006). Clonal interference as originally defined requires that beneficial mutations arise in an asexual population at sufficient frequency that they compete but at a rate low enough that two interfering mutations both arise on the same ancestral background (Gerrish & Lenski, 1998); newer theory accounts for competition between clones harboring multiple beneficial mutations Sniegowski & Gerrish, 2010). Recent estimates of beneficial mutation rates in S. cerevisiae and cancer are sufficiently high that interference between clones is expected in evolving populations (Martincorena et al., 2018;Williams et al., 2018;Wloch et al., 2001). Intriguingly, these results may also apply to cancer and infectious diseases, though precisely where these disease systems might lie on the continuum between rare and common beneficial mutations is unknown. Our findings are likely applicable to clonal interference in cancer, however. It is a fundamental property of cancers that they harbor a multitude of mutations relative to the germ line (Hanahan & Weinberg, 2000Salk et al., 2010).
In addition, cancers are heterogeneous populations of competing clones Merlo et al., 2006Merlo et al., , 2010Park et al., 2010), often constrained by a spatial structure such that cells only compete with their direct neighbors. Spatial structure can increase clonal interference depending on the population heterogeneity and migration rate (Campos et al., 2008;, though if there is little F I G U R E 2 Heat maps of relative fitness of each population on the four experimental limiting media conditions. The relative fitness values were normalized for each experimental limiting media condition to aid visual interpretation. Highest fitness on each limitation = red, lowest = blue. Boxes indicate fitness on media under which each set of populations was evolved F I G U R E 3 Relative fitness of evolved populations relative to reference strain on rich (YPD) media. Note that the relative fitness of the ancestor is centered around 1, with most evolved populations showing lower fitness on rich media than the ancestral strain, suggesting an evolutionary trade-off between adaptations to limiting conditions vs. rich media movement of cells, clones carrying different beneficial mutations may exist in the population but not come under direct competition. Additionally, in cancer, the local microenvironment may affect the selective effect of a particular clone (Karnoub et al., 2007). Since the doubling time of tumors is much slower than the doubling time of cancer cells in vitro (Friberg & Mattson, 1997), there is likely enough turnover of cells in tumors that nontoxic interventions, such as changes in nutrients and growth factors, could be used to slow the evolution of neoplastic progression.
Note that we started with a single strain that we evolved in different conditions, but have measured the fitness of the entire evolved populations, rather than cloning out strains from those populations. This provides an overall average fitness across an evolved population, but does not provide a measure of the variation in fitness among different clones within those populations. Some of that clonal variation probably generated variation in our replicates of the fitness assays (seen in Figure 1), and so was captured in our statistical tests. Tables 1 and 4 show the results of comparing these populations to the same ancestral strain.
Some results shown here may also be consistent with functional interference, a phenomenon of phenotypic interference, or tradeoff (antagonistic pleiotropy), that has also recently been linked to multi-drug resistance (Perron et al., 2012). In our experiments, functional interference could occur if mutations conferring increased fitness in one nutrient-limited component causes reduced fitness in another component. While we have attempted to rule out functional interference by choosing three nonoverlapping metabolic pathways, two results support this interpretation: 1) the growth of Glu evol and Ura evol populations, which produce negative fitnesses on limiting histidine and positive fitnesses on other environments (Figures 1 and 2) and 2) the growth of the evolved populations on rich (YPD) media, where most evolved populations show a reduced fitness compared to the ancestor (Figure 3). This suggests a trade-off between adaptation to resource limitation and growth under abundant resources.
We may also consider functional interference as a mechanism limiting the evolution of a "generalist" yeast population in the condition limiting glucose, uracil, and histidine.
We expect that the application of one vs. three selective pressures will affect the distribution of beneficial and deleterious mutations. While the total mutation rate should not change, we predicted that the three selective pressure case would create an environment in which more new mutations can have a beneficial effect, effectively skewing the distribution of the fitness effects of new mutations toward beneficial mutations and increasing the beneficial mutation supply rate. Consistent with this prediction, we see more total mutations (p = .04 for all mutations, p = .01 for nonsynonymous mutations only) that have reached fixation in the populations evolved under multiple selective pressures (Figure 4,), but consistent with clonal interference, we do not see a concomitantly increased relative fitness under three selective pressures compared to a single selective pressure (Tables 1 and 4).
It is worth noting that not all three nutrients may be limiting simultaneously over the entire course of the 500 generation evolution experiment. Rather, the fact that the AllB evol population had similar fitness to the Glu evol populations on glucose-limited media suggests that, at some point during the experiment, glucose limitation was a stronger selective pressure than histidine and uracil limitations for AllB evol . In fact, the most limiting nutrient may cycle; as a population acquires an adaptation allowing for improved growth on one limiting nutrient, it may cease to be truly limiting until adaptations are acquired for increased fitness on the other limitations. This is similar to the R* concept from ecology (Tilman, 1982).
Regardless of whether the effect is due to clonal or functional interference, we have shown that we can slow evolution by adding selective pressures to a population (Tables 1 and 2). We may be able to take advantage of this phenomenon to design new interventions for diseases and limit the spread of resistant weeds and pests in agriculture. In fact, the addition of selective pressures is entirely consistent with the ecological approaches advocated in Integrated Pest Management (Barzman et al., 2015;Kogan, 1998).
The application of the multiple selective pressures described here is a more general principle than, for example, a therapy that involves treatment with multiple bactericidal antibiotics, though this may indeed operate through a mechanism of interference (Perron

Limiting Nutrient
Fitness Test p value

| CON CLUS I ON S AND IMPLI C ATI ON S
Human diseases, including infectious diseases and cancer, undergo adaptive evolution during disease progression. Finding a way to slow adaptation could allow us to slow the rate of disease progression and the evolution of therapeutic resistance. In this study, we tested one possible mechanism for slowing the rate of evolution in a yeast (S.

TA B L E 3
Single pressure evolved populations have higher fitness on evolved limiting media than on other medias

Limiting Nutrient Fitness Test p value
Population Fitness Evolved on Limiting Media (mean ± SD)

Fitness Test
Limiting Nutrient p value Evolved Population Fitness (mean ± SD)

TA B L E 4
Most evolved populations do not have higher fitness than the ancestral strain on alternative media constraining an evolving population. In our experiment, we evolved yeast for 500 generations under either one or three selective pressures. We find that evolution under multiple selective pressures can, indeed, effectively slow the rate of adaptation.

ACK N OWLED G EM ENTS
The authors wish to thank John Koschwanez for helpful comments, John Pepper for suggesting the original idea for the study and Shane Jensen for advice on statistical analyses. We also thank the following individuals for help with maintenance of long-term cultures and counting of plates for fitness measurements: Marcin Imielinski, Rumen Kostadinov, Rachel Cohen, Sanchaika Gaur, and Erin O'Brien.

DATA AVA I L A B I L I T Y S TAT E M E N T
The yeast genomic sequence data that support the findings of this