cond-mat0302305/part1
1: 
2: \documentstyle[preprint,aps,prl,psfig]{revtex}
3: \begin{document}
4: \topmargin -0.05cm
5: \title{Origins of Chevron Rollovers in Non-Two-State Protein Folding Kinetics}
6: \author{H\"useyin K{\footnotesize{AYA}}  and Hue Sun C{\footnotesize{HAN}}}
7: \address{Protein Engineering Network of Centres of Excellence,\\
8:  Department of Biochemistry, and \\
9:  Department of Medical Genetics \& Microbiology \\
10:  Faculty of Medicine, University of Toronto \\
11:  Toronto, Ontario M5S 1A8, Canada}
12: 
13: 
14: \maketitle
15: 
16: \begin{abstract}
17: 
18: Chevron rollovers of some proteins imply that their 
19: logarithmic folding rates are nonlinear in native stability.  
20: This is predicted by lattice and continuum G\=o models to 
21: arise from diminished accessibilities of the ground state     
22: from transiently populated compact conformations under strongly
23: native conditions. Despite these models' native-centric         
24: interactions, the slowdown is due partly to kinetic trapping  
25: caused by some of the folding intermediates' nonnative 
26: topologies. Notably, simple two-state folding kinetics of     
27: small single-domain proteins are not reproduced by common 
28: G\=o-like schemes. 
29: 
30: 
31: \vskip 1cm
32: \noindent
33: {\underline {PACS Numbers}:} 87.15.Aa, 87.15.Cc, 87.15.He, 87.15.By
34: \end{abstract}
35: %===================================================================
36: \eject
37: 
38: The physical basis of protein folding is a central unresolved puzzle 
39: in molecular biology. Recently, much advance in protein folding has
40: originated from experiments on small single-domain proteins [1] with 
41: simple two-state folding and unfolding kinetics typified by that of 
42: CI2 [2], with features including: (i) single-exponential relaxation, 
43: (ii) the logarithmic folding and unfolding rates ($\ln k_{\rm f}$ 
44: and $\ln k_{\rm u}$) at constant temperature being essentially linear 
45: in chemical denaturant (urea or GuHCl) concentration, i.e., both arms 
46: of the ``chevron plot'' [3] are linear, and that (iii) the equilibrium 
47: ratio of native to denatured conformational population $K\equiv$ 
48: [native]/[denatured] $=k_{\rm f}/k_{\rm u}$. What form of intrachain 
49: interactions might give rise to such remarkable behavior is a question 
50: of fundamental biophysical interest.
51: 
52: Other proteins' corresponding properties are more complex. Often this is 
53: manifested by significant deviations [4--9] from the above linearities,
54: i.e., they exhibit chevron rollovers [10]. We refer to their kinetics
55: as non-two-state. Examples of such behavior include barnase [4,8], 
56: ribonuclease A [5], hen lysozyme [6], and U1A [7]. 
57: The present operational definition of non-two-state kinetics 
58: encompasses what some authors called ``two-state'' (though not ``simple'') 
59: when conditions (i) and (iii) above are satisfied but not (ii) [7,8].
60: Chevron rollover can also be brought about by mutation, as in S6 [7] and 
61: BPTI [9]. Thus, rather than an aberration, chevron rollover is quite 
62: ubiquitous. Therefore, ascertaining its physical origins should 
63: provide important clues to protein energetics. 
64: 
65: Chevron rollovers have been attributed to peculiarities of 
66: intermediates or transition states on postulated free energy profiles 
67: [4--8,10], or front factors' sensitivity to folding conditions [11]. 
68: Yet these phenomenological considerations do not pinpoint 
69: the physical processes involved. In this Letter, physical mechanisms 
70: underlying chevron rollovers are addressed directly by examining a 
71: multitude of trajectories from several protein chain models.
72: 
73: The recent discovery of a remarkable correlation between contact order 
74: and folding rates of simple two-state proteins [12] has led to extensive 
75: studies of G\=o-like protein models [13--17]. Hence,
76: a natural question is whether common G\=o-like constructs
77: do predict simple two-state kinetics. Somewhat surprisingly, our investigation 
78: thus far indicates that this may not be the case. Instead, chevron rollover 
79: emerges as a conspicuous feature in both lattice [11] and continuum [17]
80: G\=o models. This suggests that common native-centric [16] chain 
81: constructs can be useful for elucidating the polymer mechanisms of chevron 
82: rollovers, even though they may not be entirely adequate for simple 
83: two-state proteins. Pursuing this logic, we now analyze a thermodynamically 
84: cooperative [16] 48mer three-dimensional lattice G\=o model [14]. This model 
85: had notable impact on recent appraisals [18] of the energy 
86: landscape views of protein folding [19], but its chevron behavior 
87: has not been investigated.
88: 
89: Each native contact in this model has a favorable energy 
90: $\epsilon$ ($<0$),
91: nonnative contacts have zero energy. Folding/unfolding kinetics are modeled 
92: by Metropolis Monte Carlo (MC) dynamics with the same set of 
93: elementary chain moves as in [14]: End moves are attempted for the two 
94: chain-end monomers.  Corner and crankshaft moves are attempted for
95: other monomers with 70\% and 30\% probability respectively. Time
96: is measured by the number of attempted MC moves; $Q$ is fractional number 
97: of native contacts [15,17].
98: 
99: The chevron plots for this G\=o model and a closely related model are 
100: shown in Fig.~1. When relaxation is single-exponential (see below), 
101: $k_{\rm f}$ or $k_{\rm u}$ $=$ 1/MFPT. Most of the MFPTs 
102: here are averaged from at least 1,000 trajectories, except for a narrow
103: $\epsilon/k_{\rm B}T$ range 
104: around the transition midpoint where kinetics are relatively
105: slow (100--200 trajectories 
106: each) and for folding initiated from Fig.~2(d) (200 
107: trajectories each). In these models, the free energy $\Delta G_{\rm u}$ 
108: of unfolding 
109: to the open conformations ($Q\le 6/57$) is essentially linear in 
110: $\epsilon/k_{\rm B}T$. Thus we model denaturant concentration changes
111: by varying $\epsilon/k_{\rm B}T$ [11,20]. Adding repulsive nonnative contact
112: energies to a G\=o model [20,21] does not appear to have a significant impact 
113: on the chevron behavior. Fig.~1 shows dramatic chevron rollovers of the folding 
114: arms and very slight rollovers of the unfolding arms for both models,
115: with maximum folding rates at $\Delta G_{\rm u}/k_{\rm B}T$= 14.2 (G\=o) and 
116: 16.2 (G\=o plus repulsion). Fig.~1 indicates that deviations from simple 
117: two-state behavior can be difficult to discern under weakly native 
118: conditions [22]. To facilitate comparison with experiments, we 
119: characterize folding rollover by the difference between the hypothetical 
120: simple two-state $\ln k_{\rm f}^{\rm 2-s}$ (inclined dotted lines in 
121:  Fig.~1) and the actual (simulated) folding rate $\ln k_{\rm f}$ $\approx$ 
122: $-\ln({\rm MFPT})$ at three representative values of native stability
123: $\Delta G_{\rm u}$,
124: spanning a range typically covered by real proteins. Here, for 
125: $\Delta G_{\rm u}/k_{\rm B}T=$ (5, 10, 15), the logarithmic rollover ratio 
126: $\ln (k_{\rm f}^{\rm 2-s}/k_{\rm f})$ $=$ ($0.16$, $1.08$, $2.96$) 
127: for the G\=o model and ($0.32$, $1.36$, $3.12$) for the model with 
128: repulsive interactions. These ratios are not dissimilar to the corresponding 
129: $\ln (k_{\rm f}^{\rm 2-s}/k_{\rm f})$ $\approx$ ($0.77$, $2.42$, $4.28$) 
130: for wildtype barnase at 25$^\circ$C and pH 6.3 [4]. Under these
131: conditions, a maximum folding rate 
132: was not observed for barnase [4]. However, if a 
133: quadratic dependence [7,23] of $\ln k_{\rm f}$ vs. denaturant is assumed
134: for barnase (c.f. [8]), a maximum folding rate $\approx 230$ s$^{-1}$ may be 
135: extrapolated to occur at an hypothetical 
136: $\Delta G_{\rm u}\approx 40 k_{\rm B}T$ which is
137: much more stable than the $\Delta G_{\rm u}\approx 18.0 k_{\rm B}T$ 
138: at zero denaturant.
139: 
140: \vskip -.01cm
141: 
142:  Fig.~2(a) provides the G\=o model's conformational distributions 
143: at different native stabilities.
144: Under mildly native conditions 
145: ($\Delta G_{\rm u}<10 k_{\rm B}T$), the free energy profiles have
146: a barrier between the native and denatured minima. 
147: Under more strongly native conditions 
148: ($\Delta G_{\rm u}> 15 k_{\rm B}T$), their shapes are suggestive of
149: downhill folding [24]. The analysis of first passage time distribution 
150: [16,25] in Fig.~2(b) indicates that folding kinetics is approximately 
151: single exponential [$\ln P(t)$ linear in $t$] under mildly native
152: conditions, consistent with the 
153: observed single-exponential folding kinetics for ribonuclease A 
154: when double-jump experiments were used to eliminate the effect of 
155: {\it cis/trans} proline isomerization [5]. 
156: The behavior of wildtype barnase is
157: similar: Folding is fast and single-exponential for
158: the majority of the chains ($\approx 80\%$), the rest 
159: belongs to a slow-folding tail caused by proline isomerization [4].
160: However, when modeling conditions are strongly native (corresponding
161: conditions may not always be experimentally achievable [11]), 
162: folding kinetics is not single-exponential [circles and 
163: squares in Fig.~2(b)]. The onset of this behavior
164: occurs approximately when the $-\ln [P(Q)]$ profile becomes downhill
165: and where folding rate is maximum (c.f. Figs.~1 and 2) [11]. It would 
166: be instructive to ascertain whether 
167: this specific model prediction applies to real proteins.
168: 
169: A closer examination of the model folding trajectories indicates
170: that the slowdown leading to folding-arm chevron rolloves
171: arises from transiently populated compact non-ground-state conformations 
172: because these
173: folding ``intermediates'' have lifetimes that increase with increasingly
174: native conditions (Figs.~1, 3 and 4). Examples are shown in Fig.~3(b--d). 
175: Once one of these structures is adopted under 
176: strongly native conditions ($\epsilon/k_{\rm B}T>1.55$), it takes 
177: longer on average to reach the ground state if intrachain native 
178: contacts are more favorable (Fig.~1). Folding trajectories 
179: under strongly native conditions are qualitatively different
180: from that under milder conditions [Fig.~4(a--c)]. At the transition
181: midpoint [Fig.~4(c)], interconversions between $Q\approx 0.2$ and
182: $Q\approx 1.0$ are sudden and sharp. Relatively little time is spent at
183: intermediate $Q$ values. As $\epsilon/k_{\rm B}T$ increases, however, certain 
184: conformations with intermediate $Q\approx$ $0.6$--$0.8$ are 
185: frequented more.  
186: Even under mildly native conditions, their impeding 
187: effects on folding kinetics is already apparent from the event in Fig.~4(b)
188: depicting the chain bounces back to $Q\approx 0.2$ after achieving 
189: $Q\approx 0.8$. 
190: But the lifetimes of these ``intermediates'' are brief compared 
191: to that of the open unfolded conformations. Hence folding remains 
192: approximately single-exponential [triangles in Fig.~2(b)]. 
193: However, when conditions become more strongly native [Fig.~4(a)], 
194: some folding trajectories are dominated by ``intermediates,'' leading 
195: to a significant reduction in average 
196: folding time. But even under these circumstances it is
197: still possible to fold quickly [Fig.~4(a), inset]. 
198: Consistent with Fig.~2b (circles), this separation of time scales means 
199: that folding is no longer single-exponential.
200: 
201: In contrast to a previous report that no ``entangled misfolded state''
202: was observed during the folding of this particular G\=o model [14], 
203:  Fig.~3(b) exhibits an overall 
204: nonnative topology. For Fig.~3(c), the left side of the conformation 
205: is native, but the right side is substantially nonnative. Hence these 
206: conformations are kinetic traps in that they cannot reach the ground state 
207: without first open up somewhat by breaking some existing 
208: favorable contacts. Notwithstanding possible artifacts of lattice models 
209:  [Fig.~3(d)], this basic physical requirement rationalizes folding-arm
210: chevron rollover because favorable contacts contributing to the 
211: meta-stability of these traps are increasing difficult to break with 
212: stronger $-\epsilon/k_{\rm B}T$. 
213: 
214: This prediction appears to be robust over a range of lattice and 
215: continuum coarse-grained models [Fig.~4(d--e)] that exhibit chevron rollovers. 
216:  Fig.~4(d) shows that folding 
217: of a recent lattice model with residue-based as well as 
218: native-centric interactions are similar to that in Fig.~4(b).
219:  Fig.~4(e) shows that 
220: folding of a continuum C$_\alpha$ model 
221: under mildly and strongly [inset in (e)] native conditions are 
222: very much similar, respectively, to the corresponding
223: lattice results in Fig.~4(b) and 
224: (a). The trajectories in Fig.~4(f) from a
225: continuum model [17] with desolvation barriers [26] 
226: also show that
227: intermediate $Q$ values are more prominent during the folding process
228: when conditions are more strongly native [inset in (f)].
229: 
230: A maximum (or ``optimal'') folding rate similar to those in Fig.~1 have
231: been observed in many models (e.g., [19,20,21,23,25]) since it
232: was first noted in HP model simulations more than a decade ago [27]. This 
233: feature arises from a competition between a stronger driving force for 
234: folding and the onset of glassy dynamics under strongly native conditions
235: (see note added in proof of [20]). However, until recently [11,17],
236: the connection between this theoretical phenomenon and chevron rollover 
237: has not been recognized. Perhaps this is because the maximum folding rate 
238: often occurs near the transition midpoint for less cooperative models, 
239: and hence its relationship with chevron plots is less obvious.
240: In contrast, the models studied here possess proteinlike thermodynamic 
241: cooperativity [16]. Several basic principles now emerge: 
242: (i) Rollovers can arise from kinetic trapping [6,9,20]; but folding 
243: relaxation remains approximately single exponential
244: when trapping effects are mild [Fig.~2(b)]. (ii) We have rationalized
245: rollovers phenomenologically by front factors that depend on
246: $\Delta G_{\rm u}$ [11,17]. Physically, this dependence is 
247: likely caused by trapping and unfolding (barrier recrossing) from 
248: transiently populated compact non-ground-state conformations (Fig.~4).
249: These predictions are testable by experiments.
250: (iii) The chevron rollovers in the G\=o models presented suggest 
251: strongly that, contrary to expectation, G\=o-like pairwise additive 
252: interactions are insufficient [17] to capture the remarkable kinetics of 
253: small single-domain proteins [1]; further research is necessary to 
254: ascertain the physical origin of their simple two-state 
255: cooperativity.
256: 
257: The authors thank Walid Houry for a very helpful discussion. 
258: This work was supported in part by Canadian Institute of Health 
259: Research grant MOP-15323. 
260: 
261: 
262: %----------------------------------------------------------------------------
263: 
264: 
265: \par\vfill\eject
266: 
267: %\noindent
268: %{\large\bf References}
269: 
270: %\kern -1.5cm
271: 
272: %\footnotesize
273: \begin{thebibliography}{999}
274: 
275: \bibitem{1}
276: A. R. Fersht, Curr. Opin. Struct. Biol. {\bf 7}, 3 (1997);          
277: D. Baker, Nature {\bf 405}, 39 (2000).
278: 
279: \bibitem{2}
280: S. E. Jackson and A. R. Fersht, Biochemistry {\bf 30}, 10428 (1991).
281: 
282: \bibitem{3}
283: C. R. Matthews, Methods Enzymol. {\bf 154}, 498 (1987).
284: 
285: \bibitem{4} 
286: A. Matouschek et al., Nature {\bf 346}, 440 (1990).
287: 
288: \bibitem{5} 
289: W. A. Houry, D. M. Rothwarf and H. A. Scheraga, Nature Struct. Biol.
290: {\bf 2}, 495 (1995).
291: 
292: \bibitem{6}
293: T. Kiefhaber, Proc. Natl. Acad. Sci. USA {\bf 92}, 9029 (1995).
294: 
295: \bibitem{7} 
296: M. Silow and M. Oliveberg, Proc. Natl. Acad. Sci. USA {\bf 94}, 6084 (1997);
297: D. E. Otzen et al., Biochemistry {\bf 38}, 6499 (1999).     
298: 
299: \bibitem{8}
300: R.-A. Chu and Y. Bai, J. Mol. Biol. {\bf 315}, 759 (2002).
301: 
302: \bibitem{9}
303: R. Li, J. L. Battiste and C. Woodward, Biochemistry {\bf 41}, 2246 (2002).
304: 
305: \bibitem{10}
306: R. L. Baldwin, Fold. Des. {\bf 1}, R1 (1996).
307: 
308: \bibitem{11}
309: H. Kaya and H. S. Chan, J. Mol. Biol. {\bf 315}, 899 (2002).
310: 
311: \bibitem{12}
312: K. W. Plaxco {\it et al.}, Biochemistry {\bf 39}, 11177 (2000).
313: 
314: \bibitem{13}
315: C. Micheletti {\it et al.}, Phys. Rev. Lett. {\bf 82}, 3372 (1999).
316: 
317: \bibitem{14}
318: V. S. Pande and D. S. Rokhsar, Proc. Natl. Acad. Sci. USA {\bf 96}, 1273 (1999).
319: 
320: \bibitem{15}
321: C. Clementi, H. Nymeyer and J. N. Onuchic, J. Mol. Biol. {\bf 298}, 937 (2000).
322: 
323: \bibitem{16}
324: H. Kaya and H. S. Chan, Proteins Struct. Funct. Genet. {bf 40}, 637 (2000);
325: Phys. Rev. Lett. {\bf 85}, 4823 (2000).
326: 
327: \bibitem{17}
328: H. Kaya and H. S. Chan, J. Mol. Biol. {\bf 326}, 911 (2003).
329: 
330: \bibitem{18}
331: E. Shakhnovich and A. R. Fersht, Curr. Opin. Struct. Biol. {\bf 8}, 65 (1998);
332: V. S. Pande {\it et al.}, Curr. Opin. Struct. Biol. {\bf 8}, 68 (1998);
333: R. L. Baldwin and G. D. Rose, Trends Biochem. Sci. {\bf 24}, 77 (1999);
334: S. W. Englander, Annu. Rev. Biophys. Biomol. Struct. {\bf 29}, 213 (2000).
335: 
336: \bibitem{19}
337: J. D. Bryngelson {\it et al.}, 
338: Proteins Struct. Funct. Genet. {\bf 21}, 167 (1995);
339: K. A. Dill {\it et al.}, Protein Sci. {\bf 4}, 561 (1995); D. Thirumalai and
340: S. A. Woodson, Acc. Chem. Res. {\bf 29}, 433 (1996).
341: 
342: \bibitem{20}
343: H. S. Chan and K. A. Dill, Proteins: Struct. Funct. Genet. {\bf 30}, 2 (1998).
344: 
345: \bibitem{21}
346: M. S. Li and M. Cieplak, Eur. Phys. J. B {\bf 14}, 787 (2000).
347: 
348: \bibitem{22}
349: J. N. Onuchic {\it et al.}, Fold. Des. {\bf 1}, 441 (1996).
350: 
351: \bibitem{23}
352: N. D. Socci, J. N. Onuchic and P. G. Wolynes, J. Chem. Phys.
353: {\bf 104}, 5860 (1996).
354: 
355: \bibitem{24}
356: J. Sabelko, J. Ervin and M. Gruebele, Proc. Natl. Acad. Sci. USA
357: {\bf 96}, 6031 (1999).
358: 
359: \bibitem{25}
360: V. I. Abkevich, A. M. Gutin and E. I. Shakhnovich, J. Chem. Phys.
361: {\bf 101}, 6052 (1994).
362: 
363: \bibitem{26}
364: M. S. Cheung, A. E. Garc{\'\i}a and J. N. Onuchic, Proc. Natl. Acad. Sci.
365: USA {\bf 99}, 685 (2002).
366: 
367: \bibitem{27}
368: R. Miller {\it et al}., J. Chem. Phys. {\bf 96}, 768 (1992).
369: 
370: \end{thebibliography}
371: \par\vfill\eject
372: 
373: 
374: %----------------------------------------------------------------------------
375: 
376: 
377: \noindent
378: {\bf Fig.~1} $\quad$ MFPT is mean first passage
379: time, $k_{\rm B}T$ is Boltzmann constant times absolute temperature. 
380: Circles are for the 48mer G\=o model [14]. Squares are data from a model 
381: with the same ground-state conformation and attractive energy $\epsilon$ 
382: for each native contact but with an additional repulsive energy $-\epsilon$ 
383: for each nonnative contact (as in the HP+ model [20] and Ref.~[21]). 
384: Folding (open symbols) starts from a random self-avoiding walk, first 
385: passage is achieved when $Q=1$. Unfolding (filled symbols) starts from 
386: the $Q=1$ ground state, first passage is achieved when $Q\le 6/57$
387: because the free energy 
388: minimum of the denatured states is at $Q\approx 6/57$. Solid dashed 
389: curves are mere guide for the eye. The vertical dotted lines mark the 
390: two model's thermodynamic transition midpoints. The two pairs of V-shape 
391: dotted lines are hypothetical simple two-state chevron plots [11] based upon 
392: $\Delta G_{\rm u}$ between $Q=1$ and $Q\le 6/57$ obtained by
393: histogram techniques from sampling around the transition midpoint.
394: The triangles, asterisks, and diamonds are $-\ln({\rm MFPT})$ values for 
395: G\=o-model folding initiated (at $t=0$) respectively from the compact 
396: conformations (b), (c), and (d) in Fig.~3. Arrows indicate the 
397: $\epsilon/k_{\rm B}T$ values considered in Fig.~2. 
398: \\
399: 
400: \noindent
401: {\bf Fig.~2} $\quad$ (a) Free energy profiles for the G\=o model at the 
402: $\epsilon/k_{\rm B}T$ values indicated (c.f. Fig.~1). $P(Q)$ is Boltzmann 
403: population distribution over $Q$. [Note that $P(Q)=0$ for $Q=$ $55/57$ and 
404: $56/57$.] (b) $P(t)\Delta t$ is the fraction of folding trajectories 
405: with $t-\Delta t/2<$ first passage time $\le t+\Delta t/2$ [17], plotted 
406: in different horizontal scales for different $\epsilon/k_{\rm B}T$s 
407: [symbols as in (a)] to enhance clarity. For $\epsilon/k_{\rm B}T=$ 
408: $-1.82$, $-1.61$, $-1.47$, and $-1.28$ respectively, 2,500, 2,030, 3,500, 
409: and 1,100 trajectories are analyzed using $\Delta t/10^6=$ $30$, $1.8$, 
410: $1.6$, and $30$; the $\ln [P(t)\Delta t]$ shown are for $t$ values equal 
411: to $1$, $1/10$, $1/20$, and $1/2$ of that given by the horizontal axis. 
412: Solid and dashed lines are linear fits for $\epsilon/k_{\rm B}T=$ 
413: $-1.47$ and $-1.28$.
414: \\
415: 
416: 
417: \noindent 
418: {\bf Fig.~3} $\quad$ 
419: (a) Ground state of the G\=o model. (b--d) Transiently trapped 
420: conformations under strongly native conditions (c.f. Fig.~1),
421: with $Q=39/57$, $41/57$ and $53/57$ respectively. 
422: The dotted, dashed, dotted-dashed lines and 
423: filled circles are used to identify monomers (in all four conformations)
424: that belong to three of the straight edges and the core positions in 
425: the ground-state conformation (a). The conformation in (d) may
426: reach the ground state by a simple hinge motion of the dotted-dashed edge. 
427: However, since such a move is not available in the model [14], the chain now 
428: must first partially open up before it can access the ground state. 
429: The trapping effect of (d) is minor compared to that of (b)
430: and (c) (see Fig.~1).
431: \\
432: 
433: {\bf Fig.~4} $\quad$
434: (a--c) Folding trajectories of the 48mer G\=o model at 
435: $\epsilon/k_{\rm B}T=$ $-1.82$ (a), $-1.47$ (b), and $-1.28$ (c). The 
436: insets in (a, b) each shows a faster folding trajectories at the same 
437: given $\epsilon/k_{\rm B}T$. (d--f) Typical folding trajectories in
438: other models for comparison: (d) is from a 55mer lattice model
439: under mildly native conditions ($\epsilon/k_{\rm B}T=-1.75$) [11].
440: (e, f) are from the continuum NCS1 without-solvation (e) and 
441: with-solvation (f) Langevin dynamics models at $T=0.82$ for CI2 [17] 
442: with $\epsilon=$ $0.88$ (e) and $1.1$ (f). The insets in (e, f) show 
443: trajectories at the same $T$ but under more strongly native 
444: conditions at $\epsilon=$ $1.0$ (e) and $1.5$ (f).
445: \\
446: 
447: \end{document}
448: