Natural protein sequences are constrained by pressures to maintain required structures and functions within a complex cellular environment. However, key cellular factors shaping protein sequences (such as interactions with cellular binding partners or with the proteostasis machinery) are often unknown. To characterize functional constraints, it has been useful to determine mutational landscapes of proteins, which we define here as the effects on growth of every possible single amino acid mutation in the protein, via deep mutational scanning (Boucher et al., 2016; Fowler and Fields, 2014). Deep mutational scanning studies have provided insights into evolution of new protein functions (McLaughlin et al., 2012; Stiffler et al., 2015; Wrenbeck et al., 2017), protein design (Tinberg et al., 2013; Whitehead et al., 2012), functional trade-offs (Klesmith et al., 2017; Steinberg and Ostermeier, 2016), and adaptation to altered environments (Hietpas et al., 2013). With a few exceptions (Bandaru et al., 2017; Hietpas et al., 2013; Jiang et al., 2013; Stiffler et al., 2015), however, these studies find a general tolerance to mutation for residues outside of active sites and binding interfaces (Araya et al., 2012; Boucher et al., 2016; Klesmith et al., 2017; Roscoe et al., 2013; Wrenbeck et al., 2017) that is often explained by the absence of key environmental constraints under the selection conditions (Bandaru et al., 2017; Jiang et al., 2013; Stiffler et al., 2015).

To study the impact of multiple constraints on mutational tolerance during selection, we chose E. coli dihydrofolate reductase (DHFR) as a model system. DHFR is an essential enzyme within folate metabolism that reduces dihydrofolate to tetrahydrofolate and is necessary for thymidine production. Using this activity as the basis for an in vivo selection assay (Reynolds et al., 2011), we aimed first to measure a mutational landscape for DHFR and then to determine how a change to the cellular environment might affect the landscape. Because DHFR is known to progress through multiple conformational states during catalysis (Boehr et al., 2006; Sawaya and Kraut, 1997; Figure 1—figure supplement 1), we expected the mutational landscape of DHFR to be constrained by the requirement to adopt these different conformations. Moreover, prior work had suggested DHFR is impacted by cellular constraints such as protein quality control (Bershtein et al., 2013) and the build-up of a toxic metabolic intermediate (Schober et al., 2019). We hence expected deep mutational scanning to reveal a highly constrained mutational landscape for DHFR that would contrast with the mutational tolerance observed in other systems.

As the basis for our studies, we first sought to establish highly sensitive selection conditions for DHFR function that would be calibrated to DHFR enzymatic velocity (rate of DHF conversion per molecule of DHFR) and capable of resolving mutants with velocities near-to or faster-than wild-type. We anticipated that we would need to control DHFR protein expression (intracellular abundance) levels because two prior studies that modified the chromosomal DHFR gene had reported an overall high mutational tolerance under permissive selection conditions (Garst et al., 2017) and that DHFR abundance can be reduced to ~30% without a growth impact (Bershtein et al., 2013). We used an E. coli strain derived from ER2566 with the genes for DHFR and a downstream enzyme, thymidylate synthase, deleted in the genome and complemented on a pACYC-DUET plasmid with a weak ribosome binding site (see Materials and methods) that results in DHFR abundance at approximately 10% of the endogenous protein level (Figure 1—figure supplement 2, Figure 1—source data 1). To tightly control growth conditions, we performed selections in a turbidostat to maintain the culture in early Log phase growth (Figure 1A, Figure 1—figure supplement 3A). To quantify the effects of DHFR mutations on growth, we calculated selection coefficients (Rubin et al., 2017) from the change in allele frequency over time by deep sequencing of timepoint samples determined in biological triplicate (Figure 1B). For a panel of 14 DHFR mutants, we confirmed that the selection coefficients obtained from deep mutational scanning correlated linearly with growth rates measured separately for the individual variants in a plate reader (Figure 1—figure supplement 3B, Figure 1—source data 2), as expected. Furthermore, under our controlled selection conditions, we observed a linear relationship between selection coefficient and in vitro velocity (Figure 1C) at cytosolic substrate concentrations (Bennett et al., 2009; Kwon et al., 2008) for these DHFR mutants (Figure 1—source data 3). These results confirm that selection coefficients between −1.5 and 1.0 in our experiment are correlated with DHFR enzymatic velocity over approximately 3 orders of magnitude, and that selection can resolve mutants with higher velocities than wild-type level velocity.

Figure 1 with 7 supplements see all

Download asset Open asset

Figure 1—source data 1

Soluble DHFR expression levels in molecules per cell measured from lysate activity assays as described in Materials and methods.

The location of the DHFR gene is listed in parenthesis in the first column. Expression values corresponds to the cell strain in the column heading.

https://cdn.elifesciences.org/articles/53476/elife-53476-fig1-data1-v2.xlsx

Download elife-53476-fig1-data1-v2.xlsx

Figure 1—source data 2

Selection coefficients for –Lon selection (Figure 1—source data 1) compared to monoculture growth rates measured in a plate reader in ER2566 ∆folA/∆thyA (–Lon) as described in Materials and methods.

For values listed as ND, no detectable change in OD was measured during a 30 hr growth period.

https://cdn.elifesciences.org/articles/53476/elife-53476-fig1-data2-v2.xlsx

Download elife-53476-fig1-data2-v2.xlsx

Figure 1—source data 3

Michaelis-Menten kinetics for the set of DHFR mutants (Fierke and Benkovic, 1989; Huang et al., 1994; Reynolds et al., 2011) used to calibrate the selection are reported together with the reference from which the values were taken.

https://cdn.elifesciences.org/articles/53476/elife-53476-fig1-data3-v2.xlsx

Download elife-53476-fig1-data3-v2.xlsx

We next analyzed the deep mutational scanning data for all possible DHFR single point mutants under the calibrated selection conditions (Figure 1D, Supplementary file 1). All pairwise replicates were related with a Pearson correlation R² value of 0.70 and the median standard deviation between replicates for all selection coefficients was 0.2 (Figure 1—figure supplement 3C–E). Using this value, we defined the selection coefficient interval of 0 ±0.2 as WT-like behavior. Within this interval, the standard deviation of the selection coefficients between replicates was not correlated with changes in selection coefficient (Figure 1—figure supplement 4A). Moreover, our WT-like threshold of 0.2 was greater than the value of 0.12 for the standard deviation for wild-type synonymous codons (Figure 1—figure supplement 4B). Based on these considerations, we defined DHFR mutations with selection coefficients of <−0.2 and >0.2 as disadvantageous and advantageous, respectively. Mutations that were depleted during overnight growth (under less stringent conditions using a supplemented growth medium, see Materials and methods) were assigned a null phenotype. As expected, mutations at DHFR positions that are known to be functionally important (M20, W22, D27, L28, F31, T35, M42, L54, R57, T113, G121, D122, and S148) were generally disadvantageous or null mutations (Figure 1—figure supplement 5). These results indicate that our selection assay is a sensitive reporter of functionally important residues and that our results are consistent with previous biochemical characterization of DHFR.

In previous deep mutational scanning experiments, stringent selection typically revealed many disadvantageous mutations (Garst et al., 2017; Jiang et al., 2013; Mavor et al., 2016; Mavor et al., 2018; Stiffler et al., 2015). In contrast, the most striking observation under our conditions is the large fraction of advantageous mutations (red, Figure 1D): 736 of 3161 possible variants were advantageous (23.3%), and wild-type DHFR only ranked 1203^rd (although 467 of the 1202 higher-ranking variants fall into the WT-like interval). In direct measurements of individual growth rates under our selection conditions, the top two DHFR variants (W47L and L24V) led to increases in growth rate of 40% and 76%, respectively, when compared to wild-type DHFR (Figure 1—figure supplement 6). Advantageous mutations were widely distributed over 127 of the 159 positions of DHFR (Figure 1E). Furthermore, when we examined the DHFR structure, many of the advantageous mutations appeared to disrupt key side-chain interactions, for example by disrupting atomic packing interactions or surface salt-bridges (Figure 1—figure supplement 7).

To understand the origins of this counter-intuitive prevalence of advantageous mutations, we looked for cellular factors potentially affecting our mutational landscape. Our selection strain (Anton et al., 2016), like most standard expression strains of E. coli, is naturally deficient in Lon protease (Gur and Sauer, 2008) due to an insertion of IS186 in the lon promoter region (saiSree et al., 2001). Lon is a major component of protein quality control in E. coli (Powers et al., 2012; Sauer and Baker, 2011) responsible for degrading poorly folded proteins. Moreover, Lon had previously been implicated in degrading DHFR unstable variants in E. coli (Bershtein et al., 2013; Cho et al., 2015), and deleting Lon in an MG1655 strain of E. coli masked the deleterious impact of 2 destabilizing mutations out of a panel of 21 mutants tested in growth experiments at 30 °C (Bershtein et al., 2013). Although these 21 mutants were selected for minimal impacts on Michaelis-Menten kinetic parameters, we reasoned that the absence of Lon could be responsible for the large fraction of advantageous but potentially destabilizing mutations observed in our selection.

To test this prediction, we reintroduced chromosomal Lon expression under the control of a constitutive promoter in our selection strain, and repeated deep mutational scanning in biological triplicate (Supplementary file 2). We refer to the two regimes as +Lon and –Lon selection. The quality of +Lon selection was comparable to that of –Lon selection (Figure 2—figure supplement 1, Figure 2—figure supplement 2). Consistent with our hypothesis, the distribution of selection coefficients shifted towards more negative values in the +Lon selection, depleting positive selection coefficients and enriching for negative or null coefficients (Figure 2A). The number of advantageous mutations after reintroducing Lon decreased from 737 in –Lon selection to 384 in +Lon selection (Figure 2B), the mean selection coefficient for advantageous mutations decreased from 0.47 to 0.37, and the rank of the wild-type sequence increased by 341 to 864^th (where 479 of the 863 higher-ranked variants are in the WT-like interval) (Figure 2—figure supplement 3). The median rank of the wild-type residue over all positions decreased from eight in –Lon selection to five in +Lon selection (Figure 2—figure supplement 4).

Figure 2 with 5 supplements see all

Download asset Open asset

To examine in more detail how the mutational response of individual residues changes between selection ±Lon, we used a K-means clustering algorithm (see Materials and methods) to group all DHFR sequence positions into five categories: positions where mutations were generally advantageous (Beneficial), generally WT-like (Tolerant), variably advantageous and disadvantageous (Mixed), generally disadvantageous (Restricted), and generally null (Intolerant). Grouping was performed separately for –Lon and +Lon selection (Figure 3—source data 1). Comparing the distributions of DHFR positions in –Lon and +Lon conditions illustrates the extensive reshaping of the mutational landscape by Lon (Figure 2C,D). For –Lon selection, 28 positions (17.6%) were classified as Beneficial, where nearly every mutation was preferred over the wild-type residue. In comparison, the number of Beneficial positions decreased to 10 in +Lon selection, with only three surface-exposed positions (E48, T68, D127) common between the two Beneficial sets. Simultaneously, the number of Restricted positions increased from 42 to 67 with the reintroduction of Lon into the selection strain (Figure 2C). These results support the conclusion that Lon activity broadly penalizes mutations, including a large subset of the advantageous mutations. Overall, the changes upon modulating Lon activity lead to a model in which upregulating Lon increases constraints on DHFR, and the mutational landscape changes from being permissive when Lon is absent to being more restricted when Lon is present (Figure 2D).

To analyze the constraints imposed by Lon on the DHFR mutational landscape in structural detail, we defined a ∆selection coefficient for each amino acid residue at each position as the difference between the +Lon and –Lon selections (Figure 3A). The ∆selection coefficient values were most negative at positions in the Beneficial category and at positions with a native VILMWF or Y amino acid residue (Figure 3B, excludes Intolerant positions from –Lon selection); overall, mutations at positions with native hydrophobic residues are enriched for negative ∆selection coefficients (Figure 3—figure supplement 1A). Strikingly, the mean ∆selection coefficient was –0.71 for the 65 buried positions with <20% side-chain solvent accessible surface area, compared to –0.27 for the 79 exposed positions (Figure 3C, Figure 3—figure supplement 1B, Figure 3—source data 1). These results show that Lon has a broad impact on the mutational landscape throughout the DHFR structure but imposes particularly strong constraints in the DHFR core.

Figure 3 with 1 supplement see all

Download asset Open asset

Figure 3—source data 1

Burial classification for DHFR positions from the Getarea server (Fraczkiewicz and Braun, 1998) as described in Materials and methods.

https://cdn.elifesciences.org/articles/53476/elife-53476-fig3-data1-v2.xlsx

Download elife-53476-fig3-data1-v2.xlsx

To determine why mutations in DHFR were advantageous in the absence of Lon but less so in its presence, we selected a subset of mutations for more detailed characterization in individual experiments. We considered all positions with more than one mutation in the top 100 most advantageous mutations for the –Lon condition. We describe these positions by their location in one of four structural regions that appear to be hot-spots for the top advantageous mutations (Figure 4A,B, Figure 4—figure supplement 1): 1) exchanges between hydrophobic residues at core positions, 2) disruptions of surface residues on the beta-sheet near the active site, 3) disruptions of polar interactions with the adenine ring of NADPH, or 4) mutations to the active site or M20 loop that controls access to the active site. At these positions, we selected strongly advantageous mutations. Where possible, we selected two mutations at the same position but with significantly differing Lon sensitivities such that the set had a range of ∆selection coefficients from −0.07 to −1.46, with the exception of L24V that had a positive ∆selection coefficient. We first confirmed that the selected advantageous mutations indeed had higher cytosolic DHFR activity (the total rate of conversion of DHF to THF) in ER2566 ∆folA/∆thyA (–Lon) lysates relative to the activity for WT DHFR (Figure 4—figure supplement 2), consistent with the deep mutational scanning results.

Figure 4 with 9 supplements see all

Download asset Open asset

Figure 4—source data 1

In vitro velocity for selected advantageous measured as described in Materials and methods at multiple concentrations of DHF are reported with the standard deviation over three independent experiments.

https://cdn.elifesciences.org/articles/53476/elife-53476-fig4-data1-v2.xlsx

Download elife-53476-fig4-data1-v2.xlsx

Figure 4—source data 2

Soluble DHFR abundance levels in molecules per cell measured from lysate activity assays as described in Materials and methods.

All values are for the SMT205 plasmid transformed into the cell strain in the column heading. NM, not measured.

https://cdn.elifesciences.org/articles/53476/elife-53476-fig4-data2-v2.xlsx

Download elife-53476-fig4-data2-v2.xlsx

Figure 4—source data 3

Apparent T_m values from thermal denaturation experiments monitored by CD signal at 225 nm are reported along with the ∆selection coefficient (Lon impact) value depicted in Figure 4D.

https://cdn.elifesciences.org/articles/53476/elife-53476-fig4-data3-v2.xlsx

Download elife-53476-fig4-data3-v2.xlsx

The lysate activity assay reports on both the enzymatic activity of a DHFR variant and its intracellular abundance, [DHFR] (Bershtein et al., 2015b; Dykhuizen et al., 1987). To separate the two contributions, we purified each of the DHFR variants and determined their enzymatic velocity in vitro using concentrations of DHF that are consistent with estimates of cytosolic DHF concentration based on mass spectrometry measurements (Kwon et al., 2008). At 20 µM DHF, 16 the mutants had velocities equal and up to three-fold higher than that of WT (Figure 4C, Figure 4—figure supplement 3, Figure 4—source data 1). In contrast, the other eight mutants had velocities as much as two-fold lower than that of WT at the same DHF concentration. These results show that the higher cytosolic DHFR activity of the advantageous mutations can only partially be explained by changes in the kinetic parameters for these mutants.

We therefore examined the soluble intracellular abundance of these mutants. In the absence of Lon, we observed that mutant abundance levels varied from close-to-wild-type levels to a 20-fold increase over wild-type (Figure 4D, Figure 4—figure supplement 4, Figure 4—source data 2). Importantly, abundance decreased for most mutants in the presence of Lon (Figure 4—figure supplement 4), as expected, and these abundance decreases correspond to decreased selection coefficients (negative values in the ∆selection coefficients from Figure 3 that report on the Lon impact on selection (Figure 4—figure supplement 5)). Moreover, when considering both velocity and abundance the expected total cellular DHFR activity ([DHFR] • velocity) is increased compared to wild-type for the majority of advantageous mutants (Figure 4E, Figure 4—figure supplement 6, positions above the dotted line indicate expected cellular activity greater than wild-type). However, the expected total cellular DHFR activity is not a strong quantitative predictor of the advantageous mutants in –Lon selection (Figure 4—figure supplement 7, Figure 4—figure supplement 8). We attribute discrepancies at least in part to the difficulty of accurately quantifying rather small differences in activity and abundance, in addition to other potential complicating factors such as differential activity of cellular chaperones for different DHFR variants (Cho et al., 2015), and feedback regulation that could affect cellular concentrations of the substrate DHF (Bershtein et al., 2015a; Kwon et al., 2008). Nevertheless, our velocity and abundance measurement are in qualitative agreement with the in vivo selection. Taken together, these results suggest that increased selection coefficients arise from an interplay of effects of the mutations on cellular abundance and catalytic activity (Dykhuizen et al., 1987), and that each parameter alone is insufficient to explain the majority of the advantageous mutations. Moreover, Lon suppresses advantageous mutations at least in part by reducing their cellular abundance.

To test more directly whether advantageous mutations in DHFR destabilize the protein and whether this destabilization could explain the sensitivity to Lon expression, we measured apparent melting temperature (T_m) values from non-reversible thermal denaturation monitored by circular dichroism spectroscopy. We found that many of the advantageous mutations considerably destabilized the protein (Figure 4F, Figure 4—figure supplement 9, Figure 4—source data 3). Moreover and as expected, the ∆selection coefficients between +Lon and –Lon selection (Figure 3) are correlated with T_m (Figure 4F), except for mutations near the active site. Strikingly, when we compare different mutations at the same position, the change in ∆selection coefficients (i.e. Lon sensitivity) correlates with the change in T_m values (Figure 4G). These results indicate that the many of the selected advantageous mutations are destabilizing, and that destabilization is correlated with Lon sensitivity. One possible explanation for the selection advantage of the subset of destabilizing mutations with increased k_cat (e.g. L24V, W30F/M, M42F/Y, H114V, D116I/M, E154V) is that these mutations promote breathing motions that accelerate product release, which is rate limiting for wild-type DHFR at neutral pH (Oyen et al., 2017) and for a hyperactive DHFR mutant with a 7-fold increase in k_cat (Iwakura et al., 2006).

Taken together, our data indicate that the observed widespread changes in the mutational landscape of DHFR can be explained by a penalty for destabilizing mutations from Lon expression, leading to extensive activity – stability tradeoffs for advantageous mutations. The effect of these two selection pressures is directly observable in the structural arrangement of the mutational response categories (Figure 5, Figure 5—figure supplement 1). In –Lon conditions, mutational responses are arranged in shells around the hydride transfer site (Liu et al., 2013; Figure 5A, top), where the proportion of advantageous mutations increases with increasing distance (Figure 5B). This same spatial pattern also holds for +Lon selection (Figure 5A, bottom), but it is now superimposed with the additional pressure against destabilizing mutations such that there are no Beneficial positions in the core (Figure 5C, Figure 5—figure supplement 2). In contrast, the mutational responses as a function of distance to other DHFR sites (e.g. C5 of the NADPH adenine ring) do not show as strong of a relationship (Figure 5—figure supplement 3). These findings illustrate how the contributions from two constraints – one structural (distance from hydride transfer) and one dependent on cellular context (Lon) – can be distinguished from structural patterns in the mutational landscape.

Figure 5 with 3 supplements see all

Download asset Open asset

The naturally occurring insertion in the Lon promoter in our original selection strain, in combination with our stringent selection conditions, allowed the serendipitous discovery that advantageous mutations are remarkably prevalent throughout the DHFR structure but are also highly sensitive to Lon. The large fraction of advantageous mutations to DHFR appears to conflict with the fixation of the wild-type DHFR sequence during evolution. While Lon expression in our selection increases both the relative rank of the WT DHFR sequence (Figure 2—figure supplement 4) and the similarity between amino acid preferences from selection and from bacterial DHFR orthologues (Figure 2—figure supplement 5), there are still considerable differences: There are still 384 advantageous mutants that rank substantially better than the WT sequence even in the presence of Lon, and the amino preferences in the two selection experiments (±Lon) are more similar to each other than either is to the preferences from bacterial DHFR orthologues.

Considering these differences, we note several caveats in comparing our selection results to selection in evolution: First and most generally, screening DHFR variants under calibrated selection conditions (such as defined temperature, medium, and growth kept in early log phase) for a few generations is not expected to recapitulate the natural selection pressures on E. coli DHFR on evolutionary timescales. Second and more specifically, our selection conditions were intentionally engineered to be highly sensitive to mutations by dampening DHFR abundance to approximately 10% of the endogenous level (Figure 1—figure supplement 2). In contrast, endogenous DHFR is expected to be buffered from mutational impacts. Increasing DHFR activity or abundance in E. coli several-fold above that in wildtype strains does not increase fitness, and, conversely, reducing DHFR abundance in E. coli does not have an impact on growth until abundance is below 30% of the endogenous level (Bershtein et al., 2013; Bhattacharyya et al., 2016). Indeed, selection on mutations to the chromosomal DHFR gene did not reveal strong mutational impacts in the absence of the anti-folate drug trimethoprim (Garst et al., 2017). Third, chromosomal DHFR expression is modulated through feedback mechanism (Bershtein et al., 2015a), and it would be an interesting question how the distribution of fitness effects of DHFR mutations will be shaped by the presence of such a regulatory expression element that is absent in our selection system. Taken together, these mutational buffering effects likely explain why mutations that are advantageous in our selection are not prevalent in evolutionary DHFR sequences, and likely also explain why DHFR sequences do not vary between naturally occurring –Lon and +Lon strains of E. coli.

Nevertheless, our engineered selection conditions yielded considerable insights into constraints on mutational landscapes that are typically hidden from observation precisely because of buffering effects in natural contexts. The increase in the number of advantageous mutations in the absence of Lon shows that decreasing cellular constraints can substantially modulate the tolerance to mutation in a deep mutational scanning experiment. Because all B type E. coli strains (e.g. BL21) have the same natural Lon deficiency as our selection strain, our results could have implications for selection experiments performed in these strains over much longer time-scales such as the E. coli Long-Term Evolution Experiment (Tenaillon et al., 2016), or directed evolution strategies that often lead to mutations at positions distal to the active site.

Beyond experiments in B-type E. coli, we expect the fundamental principle of tuning trade-offs to play a role in other experimental systems. Prior work has illuminated the impact of chaperones on the effect of mutations, such as for GroEL in bacteria (Tokuriki and Tawfik, 2009) and for Hsp90 in eukaryotic cells that has been shown to buffer the phenotypic impacts of deleterious mutations (Queitsch et al., 2002). Our results highlight an opposite key role for the protein quality control machinery to tune in vivo mutational responses and lead to a model where protease activities add constraints to the mutational landscape and chaperones relieve them.

The ability to tune multiple constraints could provide a general way of controlling landscapes to drive genes into regions of sequence space that are highly responsive to external pressures. A concrete example of how this principle could be applied is in combinatorial antibiotics. Lon inactivation has been shown to increase resistance to antibiotics (Nicoloff and Andersson, 2013). Switching between compounds capable of inhibiting or activating Lon in combination with DHFR-targeting folate inhibitors such as trimethoprim could serve to variably promote destabilized resistance mutants when Lon is inhibited and then penalize those mutations when Lon is reactivated.

While the power in engineering individual gene sequences is well-recognized, we are only just beginning to explore the potential in engineering the general behavior of local sequence space. We anticipate that further study of tunable constraints will yield a new toolkit for fine control of the landscapes that guide movements through sequence space and enable unexplored engineering applications.

All plasmid and primer sequences are listed in The Appendix. Key plasmids were deposited in the Addgene plasmid repository (accession codes are listed in The Appendix). All code and python scripts are available at https://github.com/keleayon/2019_DHFR_Lon.git with key input files and example command lines (Thompson, 2020; copy archived at https://github.com/elifesciences-publications/2019_DHFR_Lon).

Source data have been provided for Figure 1 (Figure 1—source data 1–3, Supplementary file 1), Figure 2 (Supplementary files 1 and 2), Figure 3 (Figure 3—source data 1, Supplementary files 1 and 2), Figure 4 (Figure 4—source data 1–3, Supplementary file 7) and Figure 5 (Figure 3—source data 1). Code for analysis is available in our GitHub repository for this project (https://github.com/keleayon/2019_DHFR_Lon.git; copy archived at https://github.com/elifesciences-publications/2019_DHFR_Lon) along with key input files and example command lines. Raw deep sequencing data was deposited to the Sequence Read Archive in entry PRJNA590072 (BioSamples: SAMN13316587, SAMN13316662). Allele counts used to generate the selection coefficients (all figures) are reported in Supplementary files 4–6. Key plasmids (Appendix 1) will be available from Addgene.

1. 1. Thompson S

   (2019) NCBI BioProject

   ID PRJNA590072. Mapping the mutational landscape of DHFR single point mutants with perturbations to the cellular environment.

   https://www.ncbi.nlm.nih.gov/bioproject/?term=PRJNA590072

1. 1. Anton BP
   2. Fomenkov A
   3. Raleigh EA
   4. Berkmen M

   (2016) Complete Genome Sequence of the Engineered Escherichia coli SHuffle Strains and Their Wild-Type Parents

   Genome Announcements 4:16.

   https://doi.org/10.1128/genomeA.00230-16

   • Google Scholar
2. 1. Araya CL
   2. Fowler DM
   3. Chen W
   4. Muniez I
   5. Kelly JW
   6. Fields S

   (2012) A fundamental protein property, thermodynamic stability, revealed solely from large-scale measurements of protein function

   PNAS 109:16858–16863.

   https://doi.org/10.1073/pnas.1209751109

   • PubMed
   • Google Scholar
3. 1. Bandaru P
   2. Shah NH
   3. Bhattacharyya M
   4. Barton JP
   5. Kondo Y
   6. Cofsky JC
   7. Gee CL
   8. Chakraborty AK
   9. Kortemme T
   10. Ranganathan R
   11. Kuriyan J

   (2017) Deconstruction of the ras switching cycle through saturation mutagenesis

   eLife 6:e27810.

   https://doi.org/10.7554/eLife.27810

   • PubMed
   • Google Scholar
4. 1. Bennett BD
   2. Kimball EH
   3. Gao M
   4. Osterhout R
   5. Van Dien SJ
   6. Rabinowitz JD

   (2009) Absolute metabolite concentrations and implied enzyme active site occupancy in Escherichia coli

   Nature Chemical Biology 5:593–599.

   https://doi.org/10.1038/nchembio.186

   • PubMed
   • Google Scholar
5. 1. Bershtein S
   2. Mu W
   3. Serohijos AW
   4. Zhou J
   5. Shakhnovich EI

   (2013) Protein quality control acts on folding intermediates to shape the effects of mutations on organismal fitness

   Molecular Cell 49:133–144.

   https://doi.org/10.1016/j.molcel.2012.11.004

   • PubMed
   • Google Scholar
6. 1. Bershtein S
   2. Choi JM
   3. Bhattacharyya S
   4. Budnik B
   5. Shakhnovich E

   (2015a) Systems-level response to point mutations in a core metabolic enzyme modulates genotype-phenotype relationship

   Cell Reports 11:645–656.

   https://doi.org/10.1016/j.celrep.2015.03.051

   • PubMed
   • Google Scholar
7. 1. Bershtein S
   2. Serohijos AW
   3. Bhattacharyya S
   4. Manhart M
   5. Choi JM
   6. Mu W
   7. Zhou J
   8. Shakhnovich EI

   (2015b) Protein homeostasis imposes a barrier on functional integration of horizontally transferred genes in Bacteria

   PLOS Genetics 11:e1005612.

   https://doi.org/10.1371/journal.pgen.1005612

   • PubMed
   • Google Scholar
8. 1. Bhattacharyya S
   2. Bershtein S
   3. Yan J
   4. Argun T
   5. Gilson AI
   6. Trauger SA
   7. Shakhnovich EI

   (2016) Transient protein-protein interactions perturb E. coli metabolome and cause gene dosage toxicity

   eLife 5:e20309.

   https://doi.org/10.7554/eLife.20309

   • PubMed
   • Google Scholar
9. 1. Blomfield IC
   2. Vaughn V
   3. Rest RF
   4. Eisenstein BI

   (1991) Allelic exchange in Escherichia coli using the Bacillus subtilis sacB gene and a temperature-sensitive pSC101 replicon

   Molecular Microbiology 5:1447–1457.

   https://doi.org/10.1111/j.1365-2958.1991.tb00791.x

   • PubMed
   • Google Scholar
10. 1. Boehr DD
    2. McElheny D
    3. Dyson HJ
    4. Wright PE

    (2006) The dynamic energy landscape of dihydrofolate reductase catalysis

    Science 313:1638–1642.

    https://doi.org/10.1126/science.1130258

    • PubMed
    • Google Scholar
11. 1. Boucher JI
    2. Bolon DN
    3. Tawfik DS

    (2016) Quantifying and understanding the fitness effects of protein mutations: laboratory versus nature

    Protein Science 25:1219–1226.

    https://doi.org/10.1002/pro.2928

    • PubMed
    • Google Scholar
12. 1. Cho Y
    2. Zhang X
    3. Pobre KF
    4. Liu Y
    5. Powers DL
    6. Kelly JW
    7. Gierasch LM
    8. Powers ET

    (2015) Individual and collective contributions of chaperoning and degradation to protein homeostasis in E. coli

    Cell Reports 11:321–333.

    https://doi.org/10.1016/j.celrep.2015.03.018

    • PubMed
    • Google Scholar
13. 1. Dykhuizen DE
    2. Dean AM
    3. Hartl DL

    (1987)

    Metabolic flux and fitness

    Genetics 115:25–31.

    • PubMed
    • Google Scholar
14. 1. Edgar RC

    (2004) MUSCLE: multiple sequence alignment with high accuracy and high throughput

    Nucleic Acids Research 32:1792–1797.

    https://doi.org/10.1093/nar/gkh340

    • Google Scholar
15. 1. Fierke CA
    2. Benkovic SJ

    (1989) Probing the functional role of threonine-113 of Escherichia coli dihydrofolate reductase for its effect on turnover efficiency, catalysis, and binding

    Biochemistry 28:478–486.

    https://doi.org/10.1021/bi00428a011

    • PubMed
    • Google Scholar
16. 1. Fowler DM
    2. Fields S

    (2014) Deep mutational scanning: a new style of protein science

    Nature Methods 11:801–807.

    https://doi.org/10.1038/nmeth.3027

    • Google Scholar
17. 1. Fraczkiewicz R
    2. Braun W

    (1998) Exact and efficient analytical calculation of the accessible surface areas and their gradients for macromolecules

    Journal of Computational Chemistry 19:319–333.

    https://doi.org/10.1002/(SICI)1096-987X(199802)19:3<319::AID-JCC6>3.0.CO;2-W

    • Google Scholar
18. 1. Garst AD
    2. Bassalo MC
    3. Pines G
    4. Lynch SA
    5. Halweg-Edwards AL
    6. Liu R
    7. Liang L
    8. Wang Z
    9. Zeitoun R
    10. Alexander WG
    11. Gill RT

    (2017) Genome-wide mapping of mutations at single-nucleotide resolution for protein, metabolic and genome engineering

    Nature Biotechnology 35:48–55.

    https://doi.org/10.1038/nbt.3718

    • PubMed
    • Google Scholar
19. 1. Guerrero RF
    2. Scarpino SV
    3. Rodrigues JV
    4. Hartl DL
    5. Ogbunugafor CB

    (2019) Proteostasis Environment Shapes Higher-Order Epistasis Operating on Antibiotic Resistance

    Genetics 212:565–575.

    https://doi.org/10.1534/genetics.119.302138

    • Google Scholar
20. 1. Gur E
    2. Sauer RT

    (2008) Recognition of misfolded proteins by Lon, a AAA+ protease

    Genes & Development 22:2267–2277.

    https://doi.org/10.1101/gad.1670908

    • Google Scholar
21. 1. Hietpas RT
    2. Bank C
    3. Jensen JD
    4. Bolon DNA

    (2013) Shifting fitness landscapes in response to altered environments

    Evolution 67:3512–3522.

    https://doi.org/10.1111/evo.12207

    • Google Scholar
22. 1. Huang Z
    2. Wagner CR
    3. Benkovic SJ

    (1994) Nonadditivity of mutational effects at the folate binding site of Escherichia coli dihydrofolate reductase

    Biochemistry 33:11576–11585.

    https://doi.org/10.1021/bi00204a020

    • PubMed
    • Google Scholar
23. Website

    1. iGEM

    (2006) Registry of standard biological parts

    Accessed November 19, 2018.

    http://parts.igem.org/Promoters/Catalog/Anderson
24. 1. Iwakura M
    2. Maki K
    3. Takahashi H
    4. Takenawa T
    5. Yokota A
    6. Katayanagi K
    7. Kamiyama T
    8. Gekko K

    (2006) Evolutional design of a hyperactive cysteine- and methionine-free mutant of Escherichia coli dihydrofolate reductase

    Journal of Biological Chemistry 281:13234–13246.

    https://doi.org/10.1074/jbc.M508823200

    • PubMed
    • Google Scholar
25. 1. Jiang L
    2. Mishra P
    3. Hietpas RT
    4. Zeldovich KB
    5. Bolon DN

    (2013) Latent effects of Hsp90 mutants revealed at reduced expression levels

    PLOS Genetics 9:e1003600.

    https://doi.org/10.1371/journal.pgen.1003600

    • PubMed
    • Google Scholar
26. 1. Joosten RP
    2. Long F
    3. Murshudov GN
    4. Perrakis A

    (2014) The PDB_REDO server for macromolecular structure model optimization

    IUCrJ 1:213–220.

    https://doi.org/10.1107/S2052252514009324

    • PubMed
    • Google Scholar
27. 1. Klesmith JR
    2. Bacik JP
    3. Wrenbeck EE
    4. Michalczyk R
    5. Whitehead TA

    (2017) Trade-offs between enzyme fitness and solubility illuminated by deep mutational scanning

    PNAS 114:2265–2270.

    https://doi.org/10.1073/pnas.1614437114

    • PubMed
    • Google Scholar
28. 1. Kwon YK
    2. Lu W
    3. Melamud E
    4. Khanam N
    5. Bognar A
    6. Rabinowitz JD

    (2008) A domino effect in antifolate drug action in Escherichia coli

    Nature Chemical Biology 4:602–608.

    https://doi.org/10.1038/nchembio.108

    • PubMed
    • Google Scholar
29. 1. Liu CT
    2. Hanoian P
    3. French JB
    4. Pringle TH
    5. Hammes-Schiffer S
    6. Benkovic SJ

    (2013) Functional significance of evolving protein sequence in dihydrofolate reductase from Bacteria to humans

    PNAS 110:10159–10164.

    https://doi.org/10.1073/pnas.1307130110

    • PubMed
    • Google Scholar
30. 1. Magoč T
    2. Salzberg SL

    (2011) FLASH: fast length adjustment of short reads to improve genome assemblies

    Bioinformatics 27:2957–2963.

    https://doi.org/10.1093/bioinformatics/btr507

    • PubMed
    • Google Scholar
31. 1. Mavor D
    2. Barlow K
    3. Thompson S
    4. Barad BA
    5. Bonny AR
    6. Cario CL
    7. Gaskins G
    8. Liu Z
    9. Deming L
    10. Axen SD
    11. Caceres E
    12. Chen W
    13. Cuesta A
    14. Gate RE
    15. Green EM
    16. Hulce KR
    17. Ji W
    18. Kenner LR
    19. Mensa B
    20. Morinishi LS
    21. Moss SM
    22. Mravic M
    23. Muir RK
    24. Niekamp S
    25. Nnadi CI
    26. Palovcak E
    27. Poss EM
    28. Ross TD
    29. Salcedo EC
    30. See SK
    31. Subramaniam M
    32. Wong AW
    33. Li J
    34. Thorn KS
    35. Conchúir SÓ
    36. Roscoe BP
    37. Chow ED
    38. DeRisi JL
    39. Kortemme T
    40. Bolon DN
    41. Fraser JS

    (2016) Determination of ubiquitin fitness landscapes under different chemical stresses in a classroom setting

    eLife 5:e18502.

    https://doi.org/10.7554/eLife.15802

    • Google Scholar
32. 1. Mavor D
    2. Barlow KA
    3. Asarnow D
    4. Birman Y
    5. Britain D
    6. Chen W
    7. Green EM
    8. Kenner LR
    9. Mensa B
    10. Morinishi LS
    11. Nelson CA
    12. Poss EM
    13. Suresh P
    14. Tian R
    15. Arhar T
    16. Ary BE
    17. Bauer DP
    18. Bergman ID
    19. Brunetti RM
    20. Chio CM
    21. Dai SA
    22. Dickinson MS
    23. Elledge SK
    24. Helsell CVM
    25. Hendel NL
    26. Kang E
    27. Kern N
    28. Khoroshkin MS
    29. Kirkemo LL
    30. Lewis GR
    31. Lou K
    32. Marin WM
    33. Maxwell AM
    34. McTigue PF
    35. Myers-Turnbull D
    36. Nagy TL
    37. Natale AM
    38. Oltion K
    39. Pourmal S
    40. Reder GK
    41. Rettko NJ
    42. Rohweder PJ
    43. Schwarz DMC
    44. Tan SK
    45. Thomas PV
    46. Tibble RW
    47. Town JP
    48. Tsai MK
    49. Ugur FS
    50. Wassarman DR
    51. Wolff AM
    52. Wu TS
    53. Bogdanoff D
    54. Li J
    55. Thorn KS
    56. O'Conchúir S
    57. Swaney DL
    58. Chow ED
    59. Madhani HD
    60. Redding S
    61. Bolon DN
    62. Kortemme T
    63. DeRisi JL
    64. Kampmann M
    65. Fraser JS

    (2018) Extending chemical perturbations of the ubiquitin fitness landscape in a classroom setting reveals new constraints on sequence tolerance

    Biology Open 7:bio036103.

    https://doi.org/10.1242/bio.036103

    • PubMed
    • Google Scholar
33. 1. McLaughlin RN
    2. Poelwijk FJ
    3. Raman A
    4. Gosal WS
    5. Ranganathan R

    (2012) The spatial architecture of protein function and adaptation

    Nature 491:138–142.

    https://doi.org/10.1038/nature11500

    • PubMed
    • Google Scholar
34. 1. Miller GP
    2. Wahnon DC
    3. Benkovic SJ

    (2001) Interloop contacts modulate ligand cycling during catalysis by Escherichia coli dihydrofolate reductase

    Biochemistry 40:867–875.

    https://doi.org/10.1021/bi001608n

    • PubMed
    • Google Scholar
35. 1. Miller GP
    2. Benkovic SJ

    (1998) Strength of an interloop hydrogen bond determines the kinetic pathway in catalysis by Escherichia coli dihydrofolate reductase

    Biochemistry 37:6336–6342.

    https://doi.org/10.1021/bi973065w

    • PubMed
    • Google Scholar
36. 1. Nicoloff H
    2. Andersson DI

    (2013) Lon protease inactivation, or translocation of the lon gene, potentiate bacterial evolution to antibiotic resistance

    Molecular Microbiology 90:1233–1248.

    https://doi.org/10.1111/mmi.12429

    • PubMed
    • Google Scholar
37. 1. Ovchinnikov S
    2. Kamisetty H
    3. Baker D

    (2014) Robust and accurate prediction of residue-residue interactions across protein interfaces using evolutionary information

    eLife 3:e02030.

    https://doi.org/10.7554/eLife.02030

    • PubMed
    • Google Scholar
38. 1. Oyen D
    2. Fenwick RB
    3. Aoto PC
    4. Stanfield RL
    5. Wilson IA
    6. Dyson HJ
    7. Wright PE

    (2017) Defining the structural basis for allosteric product release from E. coli Dihydrofolate Reductase Using NMR Relaxation Dispersion

    Journal of the American Chemical Society 139:11233–11240.

    https://doi.org/10.1021/jacs.7b05958

    • PubMed
    • Google Scholar
39. 1. Powers ET
    2. Powers DL
    3. Gierasch LM

    (2012) FoldEco: a model for proteostasis in E. coli

    Cell Reports 1:265–276.

    https://doi.org/10.1016/j.celrep.2012.02.011

    • PubMed
    • Google Scholar
40. 1. Queitsch C
    2. Sangster TA
    3. Lindquist S

    (2002) Hsp90 as a capacitor of phenotypic variation

    Nature 417:618–624.

    https://doi.org/10.1038/nature749

    • PubMed
    • Google Scholar
41. 1. Reynolds KA
    2. McLaughlin RN
    3. Ranganathan R

    (2011) Hot spots for allosteric regulation on protein surfaces

    Cell 147:1564–1575.

    https://doi.org/10.1016/j.cell.2011.10.049

    • PubMed
    • Google Scholar
42. 1. Rodrigues JV
    2. Bershtein S
    3. Li A
    4. Lozovsky ER
    5. Hartl DL
    6. Shakhnovich EI

    (2016) Biophysical principles predict fitness landscapes of drug resistance

    PNAS 113:E1470–E1478.

    https://doi.org/10.1073/pnas.1601441113

    • PubMed
    • Google Scholar
43. 1. Roscoe BP
    2. Thayer KM
    3. Zeldovich KB
    4. Fushman D
    5. Bolon DN

    (2013) Analyses of the effects of all ubiquitin point mutants on yeast growth rate

    Journal of Molecular Biology 425:1363–1377.

    https://doi.org/10.1016/j.jmb.2013.01.032

    • PubMed
    • Google Scholar
44. 1. Rubin AF
    2. Gelman H
    3. Lucas N
    4. Bajjalieh SM
    5. Papenfuss AT
    6. Speed TP
    7. Fowler DM

    (2017) A statistical framework for analyzing deep mutational scanning data

    Genome Biology 18:150.

    https://doi.org/10.1186/s13059-017-1272-5

    • PubMed
    • Google Scholar
45. 1. saiSree L
    2. Reddy M
    3. Gowrishankar J

    (2001) IS186 insertion at a hot spot in the lon promoter as a basis for lon protease deficiency of Escherichia coli B: identification of a consensus target sequence for IS186 transposition

    Journal of Bacteriology 183:6943–6946.

    https://doi.org/10.1128/JB.183.23.6943-6946.2001

    • PubMed
    • Google Scholar
46. 1. Salis HM
    2. Mirsky EA
    3. Voigt CA

    (2009) Automated design of synthetic ribosome binding sites to control protein expression

    Nature Biotechnology 27:946–950.

    https://doi.org/10.1038/nbt.1568

    • PubMed
    • Google Scholar
47. 1. Sanner MF
    2. Olson AJ
    3. Spehner JC

    (1996) Reduced surface: an efficient way to compute molecular surfaces

    Biopolymers 38:305–320.

    https://doi.org/10.1002/(SICI)1097-0282(199603)38:3<305::AID-BIP4>3.0.CO;2-Y

    • PubMed
    • Google Scholar
48. 1. Sauer RT
    2. Baker TA

    (2011) AAA+ proteases: atp-fueled machines of protein destruction

    Annual Review of Biochemistry 80:587–612.

    https://doi.org/10.1146/annurev-biochem-060408-172623

    • PubMed
    • Google Scholar
49. 1. Sawaya MR
    2. Kraut J

    (1997) Loop and subdomain movements in the mechanism of Escherichia coli dihydrofolate reductase: crystallographic evidence

    Biochemistry 36:586–603.

    https://doi.org/10.1021/bi962337c

    • PubMed
    • Google Scholar
50. 1. Schindelin J
    2. Arganda-Carreras I
    3. Frise E
    4. Kaynig V
    5. Longair M
    6. Pietzsch T
    7. Preibisch S
    8. Rueden C
    9. Saalfeld S
    10. Schmid B
    11. Tinevez JY
    12. White DJ
    13. Hartenstein V
    14. Eliceiri K
    15. Tomancak P
    16. Cardona A

    (2012) Fiji: an open-source platform for biological-image analysis

    Nature Methods 9:676–682.

    https://doi.org/10.1038/nmeth.2019

    • PubMed
    • Google Scholar
51. 1. Schober AF
    2. Mathis AD
    3. Ingle C
    4. Park JO
    5. Chen L
    6. Rabinowitz JD
    7. Junier I
    8. Rivoire O
    9. Reynolds KA

    (2019) A Two-Enzyme adaptive unit within bacterial folate metabolism

    Cell Reports 27:3359–3370.

    https://doi.org/10.1016/j.celrep.2019.05.030

    • PubMed
    • Google Scholar
52. 1. Smith CA
    2. Shi CA
    3. Chroust MK
    4. Bliska TE
    5. Kelly MJS
    6. Jacobson MP
    7. Kortemme T

    (2013) Design of a phosphorylatable PDZ domain with peptide-specific affinity changes

    Structure 21:54–64.

    https://doi.org/10.1016/j.str.2012.10.007

    • PubMed
    • Google Scholar
53. 1. Steinberg B
    2. Ostermeier M

    (2016) Shifting fitness and epistatic landscapes reflect Trade-offs along an evolutionary pathway

    Journal of Molecular Biology 428:2730–2743.

    https://doi.org/10.1016/j.jmb.2016.04.033

    • PubMed
    • Google Scholar
54. 1. Stiffler MA
    2. Hekstra DR
    3. Ranganathan R

    (2015) Evolvability as a function of purifying selection in TEM-1 β-lactamase

    Cell 160:882–892.

    https://doi.org/10.1016/j.cell.2015.01.035

    • PubMed
    • Google Scholar
55. 1. Tenaillon O
    2. Barrick JE
    3. Ribeck N
    4. Deatherage DE
    5. Blanchard JL
    6. Dasgupta A
    7. Wu GC
    8. Wielgoss S
    9. Cruveiller S
    10. Médigue C
    11. Schneider D
    12. Lenski RE

    (2016) Tempo and mode of genome evolution in a 50,000-generation experiment

    Nature 536:165–170.

    https://doi.org/10.1038/nature18959

    • PubMed
    • Google Scholar
56. 1. Thomason LC
    2. Sawitzke JA
    3. Li X
    4. Costantino N
    5. Court DL

    (2014) Recombineering: genetic engineering in Bacteria using homologous recombination

    Current Protocols in Molecular Biology 106:1.16–1-39.

    https://doi.org/10.1002/0471142727.mb0116s106

    • Google Scholar
57. Software

    1. Thompson S

    (2020) 2019_DHFR_Lon, version c3e2201

    GitHub.

    https://github.com/keleayon/2019_DHFR_Lon
58. 1. Tinberg CE
    2. Khare SD
    3. Dou J
    4. Doyle L
    5. Nelson JW
    6. Schena A
    7. Jankowski W
    8. Kalodimos CG
    9. Johnsson K
    10. Stoddard BL
    11. Baker D

    (2013) Computational design of ligand-binding proteins with high affinity and selectivity

    Nature 501:212–216.

    https://doi.org/10.1038/nature12443

    • PubMed
    • Google Scholar
59. 1. Tokuriki N
    2. Tawfik DS

    (2009) Chaperonin overexpression promotes genetic variation and enzyme evolution

    Nature 459:668–673.

    https://doi.org/10.1038/nature08009

    • PubMed
    • Google Scholar
60. 1. Whitehead TA
    2. Chevalier A
    3. Song Y
    4. Dreyfus C
    5. Fleishman SJ
    6. De Mattos C
    7. Myers CA
    8. Kamisetty H
    9. Blair P
    10. Wilson IA
    11. Baker D

    (2012) Optimization of affinity, specificity and function of designed influenza inhibitors using deep sequencing

    Nature Biotechnology 30:543–548.

    https://doi.org/10.1038/nbt.2214

    • PubMed
    • Google Scholar
61. 1. Wrenbeck EE
    2. Azouz LR
    3. Whitehead TA

    (2017) Single-mutation fitness landscapes for an enzyme on multiple substrates reveal specificity is globally encoded

    Nature Communications 8:15695.

    https://doi.org/10.1038/ncomms15695

    • PubMed
    • Google Scholar

Author details

1. Samuel Thompson

   Graduate Group in Biophysics, University of California San Francisco, San Francisco, United States

   Contribution

   Conceptualization, Software, Formal analysis, Funding acquisition, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing

   For correspondence

   Thompson.SamuelM@gmail.com

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-6468-9538

2. Yang Zhang

   Department of Bioengineering and Therapeutic Sciences, University of California San Francisco, San Francisco, United States

   Contribution

   Formal analysis, Investigation, Writing - review and editing

   Competing interests

   No competing interests declared

3. Christine Ingle

   The Green Center for Systems Biology, University of Texas Southwestern Medical Center, Dallas, United States

   Contribution

   Investigation, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-0203-2845

4. Kimberly A Reynolds

   1. The Green Center for Systems Biology, University of Texas Southwestern Medical Center, Dallas, United States
   2. Department of Biophysics, University of Texas Southwestern Medical Center, Dallas, United States

   Contribution

   Resources, Formal analysis, Supervision, Funding acquisition, Project administration, Writing - review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-4805-0317

5. Tanja Kortemme

   1. Graduate Group in Biophysics, University of California San Francisco, San Francisco, United States
   2. Department of Bioengineering and Therapeutic Sciences, University of California San Francisco, San Francisco, United States
   3. Chan Zuckerberg Biohub, San Francisco, United States

   Contribution

   Conceptualization, Resources, Formal analysis, Supervision, Funding acquisition, Writing - original draft, Project administration, Writing - review and editing

   For correspondence

   kortemme@cgl.ucsf.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-8494-680X

• 2,571

  views

• 298

  downloads

• 22

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.