Direct Demonstration of an Adaptive Constraint

See allHide authors and affiliations

Science  20 Oct 2006:
Vol. 314, Issue 5798, pp. 458-461
DOI: 10.1126/science.1133479


The role of constraint in adaptive evolution is an open question. Directed evolution of an engineered β-isopropylmalate dehydrogenase (IMDH), with coenzyme specificity switched from nicotinamide adenine dinucleotide (NAD) to nicotinamide adenine dinucleotide phosphate (NADP), always produces mutants with lower affinities for NADP. This result is the correlated response to selection for relief from inhibition by NADPH (the reduced form of NADP) expected of an adaptive landscape subject to three enzymatic constraints: an upper limit to the rate of maximum turnover (kcat), a correlation in NADP and NADPH affinities, and a trade-off between NAD and NADP usage. Two additional constraints, high intracellular NADPH abundance and the cost of compensatory protein synthesis, have ensured the conserved use of NAD by IMDH throughout evolution. Our results show that selective mechanisms and evolutionary constraints are to be understood in terms of underlying adaptive landscapes.

The old notion of natural selection as an omnipotent force in biological evolution has given way to one where adaptive processes are constrained by physical, chemical, and biological exigencies (14). Whether constraint and/or stabilizing selection explain phenotypic stasis, in the fossil record and in phylogenies, remains an open question (5). Direct experimental tests of constraint are scarce (69). Even tight correlations among traits, at once suggestive of constraint, can be broken by artificial selection to produce new phenotypic combinations (8, 9). Despite all circumstantial evidence, results from direct experimental tests imply that selection is largely unconstrained.

The direct experimental test for constraint is conceptually simple. A phenotype is subjected to selection (natural or artificial) in an attempt to break the postulated constraint (69). A response to selection indicates a lack of constraint. No response to selection indicates the presence of a constraint. However, the cause of a constraint is rarely specified because the etiologies of most phenotypes are not well understood, their relationships to fitness are usually opaque, and a lack of response to selection may reflect nothing more than a lack of heritable variation (4, 7). If the cause of a constraint is to be elucidated, it must be for a simple phenotype whose relationship to fitness is understood.

Coenzyme use by β-isopropylmalate dehydrogenase (IMDH) is a simple phenotype whose etiology and relationship to fitness are understood (10, 11). IMDHs catalyze the oxidative decarboxylation of β-isopropylmalate to α-ketoisocaproate during the biosynthesis of leucine, an essential amino acid. All IMDHs use nicotinamide adenine dinucleotide (NAD) as a coenzyme (cosubstrate). This invariance of function among IMDHs hints at the presence of ancient constraints, even though some related isocitrate dehydrogenases (IDHs) use NADP instead (12, 13).

Structural comparisons with related NADP-using IDHs identify amino acids controlling co-enzyme use (1416) (Fig. 1A). Introducing five replacements (Asp236 → Arg, Asp289 → Lys, Ile290 → Tyr, Ala296 → Val, and Gly337 → Tyr) into the coenzyme-binding pocket of Escherichia coli leuB–encoded IMDH by site-directed mutagenesis causes a complete reversal in specificity (10, 11): NAD performance (Math, where Km is the Michaelis constant) is reduced by a factor of 340, from 68 × 103 M–1 s–1 to 0.2 × 103 M–1 s–1, whereas NADP performance (Math) is increased by a factor of 70, from 0.49 × 103 M–1 s–1 to 34 × 103 M–1 s–1. The engineered LeuB[RKYVYR] mutant (the final R represents Arg341, already present in wild-type E. coli IMDH) is as active and as specific toward NADP as the wild-type enzyme is toward NAD. Evidently, protein architecture has not constrained IMDH to use NAD since the last common ancestor.

Fig. 1.

The molecular anatomy of an adaptive constraint. (A) Structural alignment of the coenzyme binding pockets of E. coli IMDH (14) (brown main chain) with bound NAD modeled from Thermus thermophilus IMDH (15) and E. coli IDH (16) (green main chain) with bound NADP. Only key residues are shown (gray, carbon; red, oxygen; blue, nitrogen; yellow, phosphorus) with labels designating the amino acid and site number in IMDH followed by the amino acid in IDH. Coenzyme use is determined by H bonds to NAD (brown lines) and to the 2′-phosphate (2′P) of NADP (green lines). (B) Mutations in the coenzyme binding pocket that stabilize NADP also stabilize NADPH. The ensuing correlation in affinities for NADP (Embedded Image) and NADPH (Embedded Image) seen with engineered mutants (circles) (11) is retained in the screened mutants (dots). (C) The adaptive landscape for coenzyme use by IMDH (11) showing the phenotype-fitness map (blue surface) determined in chemostat competition and described by equation S5 (11, 18) and the predicted distribution of performances for 512 mutants in the coenzyme binding pocket (gray dots), with wild type (red dot), LeuB[RKYVYR] (black dot), and single amino acid replacements in LeuB[RKYVYR] coenzyme binding pocket (pink dots). The fitness of LeuB[RKYVYR] is hypothesized to be lower than in the wild type because of strong inhibition by abundant intracellular NADPH (11).

Despite similar in vitro performances, the NADP-specific LeuB[RKYVYR] mutant is less fit than the NAD-specific wild type (11). IMDHs, wild-type and mutant alike, display a factor of 30 higher affinity for the reduced form of NADP (NADPH) (coproduct) than for NADP (Fig. 1B). We suggested the LeuB[RKYVYR] mutant is subject to intense inhibition by intracellular NADPH, which is far more abundant in vivo than is NADP (11, 17). The inhibition slows leucine biosynthesis, reduces growth rate, and lowers Darwinian fitness (Fig. 1C). The wild type retains high fitness because its affinity for NADPH is low, whereas inhibition by NADH, which is far less abundant than NAD in vivo, is ineffective. Perhaps as a consequence of differences in Michaelis complex structure (18), IDH is not subject to such intense NADPH inhibition and hence could evolve NADP use.

We hypothesize that IMDH is constrained to use NAD because the strong inhibition associated with NADP use reduces fitness. However, identifying the mechanism of selection (NADPH inhibition) is not synonymous with identifying the causes of constraint. Mutations that increase Math (maximum rate of NADP turnover), that break the correlation in NADP and NADPH affinities (Fig. 1B), or that eliminate the trade-off in coenzyme performances (Fig. 1C) could each benefit the LeuB[RKYVYR] mutant without compromising its performance with NADP (18). We therefore hypothesize that IMDH is constrained to use NAD by three causes: an upper limit to Math, an unbreakable correlation in the affinities of NADP and NADPH, and an inescapable trade-off in coenzyme performance.

We used directed evolution (targeted random mutagenesis and selective screening) (1921) to test whether NADP-specific IMDHs with higher fitness could be isolated. Random substitutions were introduced into leuB[RKYVYR] by means of error-prone polymerase chain reaction (18). Mutated alleles were ligated downstream of the T7 promoter in pETcoco (a stable single-copy vector) and transformed into strain RFS[DE3] (leuA+BamC+, with T7 RNA polymerase expressed from a chromosomal lacUV5 promoter). Sequencing unscreened plasmids revealed that, on average, each 1110–base pair leuB[RKYVYR] received two nucleotide substitutions. From the pattern of base substitutions in these mutants and assuming Poisson statistics, we estimate that only 10.5 (0.4%) of the 2431 possible amino acid replacements were missing in our mutant library (18).

Our experimental design used decreased IMDH expression to provide a simple selective screen for mutations in leuB[RKYVYR] that increase growth and/or fitness. As predicted from the phenotype-fitness map (Fig. 2A), selection against leuB[RKYVYR] intensified as IMDH expression was lowered in the presence of excess glucose (Fig. 2B). Using 40 μM isopropyl-β-d-thiogalactopyranoside (IPTG) to induce a low level of IMDH expression, we found that cells harboring leuB[WT] formed large colonies at 24 hours, whereas cells harboring leuB[RKYVYR] barely formed pinprick colonies at 48 hours. We chose colony formation at 48 hours as the least stringent criterion compatible with reliably identifying beneficial mutations in leuB[RKYVYR]. Longer periods of growth (or higher concentrations of IPTG) allowed unmutated leuB[RKYVYR] to form colonies, whereas shorter periods (or lower concentrations of IPTG) produced no colonies.

Fig. 2.

The phenotypic basis of the genetic screen. Lowered expression in the presence of excess glucose brings IMDH to saturation with isopropylmalate, increasing coenzyme affinities. (A) Lowering IMDH expression in the chemostat-derived adaptive landscape [lower concentration of E in equation S5 (11, 18)] is predicted to reduce the fitness of LeuB[RKYVYR] (white sphere) far more than that of the wild type (red sphere). (B) IPTG-controlled expression of IMDHs ligated downstream of the T7 promoter in pETcoco in strain RFS[DE3] (leuA+BamC+, with T7 RNA polymerase expressed from a chromosomal lacUV5 promoter) confirms that lower expression affects growth in minimal glucose medium of the LeuB[RKYVYR] (white) more than in the wild type (red). Plasmids lacking LeuB (black) are incapable of growth except in the presence of leucine.

Of the 100,000 mutated plasmids screened, 134 (representing 107 distinct isolates) formed colonies within 48 hours (table S1). Each had either a substitution in the 5′ leader sequence upstream of leuB[RKYVYR], an amino acid replacement in the coenzyme binding pocket, or both. Upstream substitutions occurred at three nucleotide positions: –3, –9, and –14 relative to the leuB AUG start codon (fig. S1). Ten amino acid replacements were found at three codons in the coenzyme binding pocket.

At first glance, the positive response to selection might suggest that coenzyme use by IMDH is unconstrained. Upstream substitutions in the Shine-Dalgarno sequence (positions –9 and –14), as well as a new AUG start codon (position –3) that replaces the less efficient GUG start codon, presumably derive their benefits through increases in expression because their kinetics are unchanged. Unexpectedly, however, beneficial amino acid replacements in the coenzyme binding pocket eliminate H bonds to the 2′-phosphate of NADP to cause striking reductions in NADP performance (Table 1). Although some mutants have improved NAD performance, others remain unchanged and several show reduced NAD performance. Isolated amino acid replacements outside the coenzyme binding pocket, which might have been expected to increase Math or to break the correlation in affinities between NADP and NADPH (Math and Math), have no detectable functional effects (table S2, those associated with beneficial 5′-leader mutations in Table 1). No doubt they, along with 126 silent substitutions (table S1), hitchhiked through the genetic screen with the beneficial mutations.

Table 1.

Kinetic effects of amino acid replacements in LeuB[RKYVYR] isolated at 48 hours growth. Standard errors are <13% of estimates. Single-letter abbreviations for amino acid residues: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; Y, Tyr.

Performance (A)Performance (B)Preference (A/B)KiNADPH (μM)
Km (μM) kcat (s-1) kcat/Km (mM-1 s-1) Km (μM) kcat (s-1) kcat/Km (mM-1 s-1)
DDIAGR (wild type) 8400 4.10 0.4900 101 6.90 68.3200 0.007 254.30
RKYVYR 183 6.20 33.8800 4108 0.83 0.2000 169.400 9.90
Beneficial replacements
K235N 3919 5.95 1.5200 6356 0.50 0.0800 19.000 129.00
K235N,-3(M),A60V 1744 5.06 2.9000 5395 0.44 0.0800 35.700 56.00
K235R 554 6.39 11.5000 3292 0.36 0.1100 106.100 26.00
K235R,E63V 658 6.75 10.3000 3748 0.46 0.1200 84.100 28.00
K289EView inline 3449 2.71 0.7900 4897 1.98 0.4000 2.000 133.50
K289EView inline,D317E 3554 3.62 1.0200 5580 2.09 0.3700 2.800 112.00
K289EView inline,D87Y,S182F 2551 2.54 1.0000 5750 3.65 0.6300 1.600 105.20
K289EView inline,F102L 3891 3.66 0.9400 4823 1.10 0.2300 4.100 131.90
K289M 2216 4.13 1.8600 4034 0.48 0.1200 15.500 78.60
K289M,F170L 1945 2.50 1.2900 2947 0.48 0.1600 8.100 87.20
K289NView inline 1141 6.32 5.5400 4997 1.14 0.2300 24.300 44.00
K289NView inline,R187H 1146 5.83 5.0900 3069 0.79 0.2600 19.600 44.70
K289NView inline,H367Q 1248 5.27 4.2200 4555 1.05 0.2300 18.300 57.80
K289T 1596 6.18 3.8700 4056 0.78 0.1900 20.400 71.60
K289T,Q157H 1696 5.66 3.3400 5038 0.93 0.1800 18.600 72.60
Y290C 988 5.21 5.2700 3792 1.02 0.2700 19.600 49.00
Y290C,R152C 910 3.15 3.4600 3844 0.90 0.2300 14.800 43.00
Y290D 10213 3.13 0.3100 9040 0.82 0.0900 3.400 344.00
Y290FView inline 1658 5.59 3.3700 3110 0.65 0.2100 16.000 59.70
Y290F,P97S,D314E 1097 4.64 4.2300 4158 0.71 0.1700 24.800 57.40
Y290N,N52T 2381 3.44 1.4400 10660 0.26 0.0200 59.500 90.00
Replacements with beneficial 5′-leader mutations
RKYVYR 183 6.20 33.8800 4108 0.83 0.2000 169.400 9.90
-3(M)View inline,E66D 208 2.97 14.2800 5002 0.30 0.0600 238.000 11.90
-3(M)View inline,E66D,E331D 267 7.99 29.9000 3403 0.34 0.1000 296.900 14.00
E82D 221 5.44 24.6200 4735 0.35 0.0700 351.700 14.10
K100R,A229T,L248M 137 5.58 40.3700 4079 0.45 0.1100 370.300 11.50
G156R,K289R,Y311H 292 5.40 18.4900 4464 0.54 0.1200 144.800 17.40
E173D 219 6.39 29.1800 5304 0.89 0.1700 171.600 6.60
Y337N,G131 171 5.71 33.3900 3367 0.24 0.0700 468.500 8.00
Y337H,P85S,E181K 318 5.21 16.3800 2409 0.22 0.0900 183.600 12.80
  • View inline* Switch from NADP to NAD requires two base substitutions in one codon. This is a possible transitional amino acid produced by a single base substitution (11).

  • View inline The -3(M) designates a G to A substitution at base -3 that creates a new start codon.

  • Our results suggest that increases in expression are beneficial, whereas increases in NADP performance are not. This seeming paradox is resolved if there are no mutations capable of breaking the upper limit to Math, the correlation in affinities for NADP and NADPH, or the trade-off in co-enzyme performance. With these constraints, and with reduced expression in the genetic screen, the phenotype-fitness map near LeuB[RKYVYR] remains flat with respect to increases in NADP performance (Fig. 2A). By contrast, severe losses of NADP performance are predicted to be beneficial as correlated reductions in the affinities for NADPH free up IMDH for use with abundant NAD. As predicted, all beneficial LeuB[RKYVYR] mutants have reduced affinities for both NADP and NADPH (Table 1). Unaffected by constraints, increases in expression are unconditionally beneficial (18). These results support the hypothesis that NADP-specific IMDHs function poorly in vivo because of strong inhibition by abundant NADPH. That reductions in NADP performance and increases in expression are both beneficial are the predicted consequences of a phenotype-fitness map constrained by an upper limit to Math, a correlation in affinities for NADP and NADPH, and a trade-off in coenzyme performance.

    Breaking any one constraint would allow NADP-specific IMDHs to evolve. Yet no mutant increases Math, no mutant uncouples the affinities for NADP and NADPH, and no mutant breaks the trade-off in coenzyme performance. These conclusions are not the result of a selective screen that is too stringent. Of 35 colonies, representing 26 distinct mutants, that appeared after the 48-hour limit, 17 had no nucleotide substitutions in the 5′ leader sequence or amino acid replacements in LeuB[RKYVYR] (table S3). Amino acid replacements in the other mutants neither improved enzyme performance nor decreased NADPH inhibition (table S4). That no additional beneficial mutations were recovered with relaxed criteria demonstrates that the selective screen was not overly stringent.

    Nor are the results a consequence of inadequate sampling of protein sequence space. Natural adaptive evolution fixes advantageous mutations sequentially (2224). Indeed, experimental evolution demonstrates that advantageous double mutants in the evolved β-galactosidase of E. coli are not evolutionarily accessible and perforce must be accumulated as sequential advantageous mutations (25). Hence, screening all single substitutions is a sufficient sampling of protein sequence space for robust evolutionary conclusions. We estimate that only 0.04 advantageous amino acid replacements are missing from the mutant leuB[RKYVYR] library (18). We conclude that mutations capable of breaking the limit, the correlation, or the trade-off are unlikely to ever be fixed in populations because they are exceedingly rare (they may not exist), because they are minimally advantageous, or both.

    The two remaining ways to evolve an NADP-specific IMDH are to reduce intracellular NADPH pool and, as our results show, to increase expression. Reducing intracellular NADPH relieves the inhibition but, as experiments deleting sources of NADPH show (13), the disruption to the rest of metabolism costs far more than the benefit to be gained. The phenotype-fitness map (Fig. 1C) imposes a law of diminishing returns such that LeuB[RKYVYR] must be expressed above wild-type levels by a factor of 100 to overcome the inhibition by NADPH (18). Diverting resources away from other metabolic needs toward compensatory protein synthesis would impose a protein burden (2629) sufficient to prevent the evolution of NADP-specific IMDHs.

    The production of unnatural phenotypes, by artificial selection or molecular engineering, is not sufficient to conclude that evolutionary constraints are absent entirely. Rather, potential constraints underlying a conserved phenotype can be identified from the relationships among genotype, phenotype, and fitness that define an adaptive landscape. Experimental evolution can then be used to test their existence. Using this approach, we have shown how certain structure-function relationships in IMDH have constrained its coenzyme phenotype since the last common ancestor.

    Supporting Online Material

    Materials and Methods

    SOM Text

    Figs. S1 to S3

    Tables S1 to S5


    References and Notes

    View Abstract

    Navigate This Article