Crystal structures of MBP fusion proteins

Abstract

Although chaperone‐assisted protein crystallization remains a comparatively rare undertaking, the number of crystal structures of polypeptides fused to maltose‐binding protein (MBP) that have been deposited in the Protein Data Bank (PDB) has grown dramatically during the past decade. Altogether, 102 fusion protein structures were detected by Basic Local Alignment Search Tool (BLAST) analysis. Collectively, these structures comprise a range of sizes, space groups, and resolutions that are typical of the PDB as a whole. While most of these MBP fusion proteins were equipped with short inter‐domain linkers to increase their rigidity, fusion proteins with long linkers have also been crystallized. In some cases, surface entropy reduction mutations in MBP appear to have facilitated the formation of crystals. A comparison of the structures of fused and unfused proteins, where both are available, reveals that MBP‐mediated structural distortions are very rare.

MBP as a Crystallization Chaperone

In 2003, Smyth et al. reviewed the crystal structures of fusion proteins with large affinity tags.1 At that time, the list contained one glutathione S‐transferase (GST), one thioredoxin (TRX), and four maltose‐binding protein (MBP) fusions. Among these six reported fusion protein structures, only three (all of them MBP fusions) were ever deposited in the PDB (1MG1, 1HSJ, 1MH3). Presently, over a decade later, there are still only three TRX (3DXB, 4KCA, 4KCB) and six GST (1B8X, 1BG5, 1DUG, 1GNE, 3QMZ, 4AI6) fusion protein structures in the PDB, but the number of MBP fusion protein structures has skyrocketed to more than 100 (Fig. 1; Table 1). It therefore seems like an opportune time to examine the large number of MBP fusion structures in some detail and see what lessons can be drawn from them.

Figure 1.

Cumulative number of MBP fusion protein structures deposited in the PDB between 1999 and 2015.

Table 1.

Structures of MBP Fusion Proteins

PDB	Protein	Organism	Length	Linker sequence	RES.	YR.	S.G.	REF.
1Y4C	Designed helical protein	NA	108	DEALKDAQTNSSSNNNNNNNNNLGIEGRSSEEL	1.9	2004	P212121	6
4O4B	Inositol hexakisphosphate kinase EhIP6KA	E. hystolytica	239	DEALKDAQTNSGSDITSLYKKAGSAAAVLEENLYFQGSFTDGQYL	1.8	2013	P212121	16
1YTV	V1 vasopressin receptor fragment	H. sapiens	59	DEALKDAQTNSSSNNNNNNNNNNLGIEENLYFQGSSFPCC	1.8	2005	P21	23
4JKM	Beta‐glucuronidase	C. perfringens	599	DEALKDAQTNSSSNNNNNNNNNNRDLGTENLYFQSNAMLYPI	2.3	2013	C2221	20
3HST	RNase H domain	M. tuberculosis	141	DEALKDAQTNSSSNNNNNNNNNNLGIEGRSVKVV	2.2	2009	P21	36
3OAI	Myelin protein extracellular domain	H. sapiens	115	DEALKDAQTNNNNNNNNNNNNNNNNNNNNIVVYT	2.1	2010	P21	11
1T0K	L30/RNA complex	S. cerevisiae	105	DEALKDAQTNSSSVPGRGSIEGRMAPVK	3.2	2004	P41212	5
1NMU	L30	S. cerevisiae	104	DEALKDAQTNSSSVPGRGSIEGRAAPVKS	2.3	2003	P212121	3
3A3C	Tim40 core domain	S. cerevisiae	73	DEALKDAQTNSSSVPGRGSIEGRPEFAY	2.5	2009	P212121	27
4GIZ	E6	HPV16	142	DAALAAAQTNAAAELTLQELLGEERFQDPQ	2.5	2012	P212121	42
4IRL	GBP‐NLRP1 CARD domain	D. rerio	95	DAALAAAQTNAARAAAASEFVD	1.5	2013	P212121	13
3MQ9	Tetherin	H. sapiens	88	DEALKDAQTRIT/AARDG	2.8	2010	P21	39
4GLI	Survival motor neuron protein	H. sapiens	32	DEALKDAQTRIT/MLISW	1.9	2012	C2	47
4PQK	RepA	S. aureus	120	DEALKDAQT/HNFDD	3.4	2014	P1	55
4B3N	TRIM5alpha	M. mulatta	202	DEALKDAQTRIT/RRVFR	3.3	2012	C2	45
3LBS	PRR‐IC domain	H. sapiens	20	DEALKDA/ADPGYD	2.2	2010	P212121	10
3LC8	PRR‐IC domain	H. sapiens	20	DEALKDA/ADPGYD	2.0	2010	P212121	10
4R0Y	GKAP GH1 domain	R. norvegicus	140	DEALAMDGHWFL	2.0	2014	P21212	57
2OK2	MutS C‐domain	E. coli	31	DEALKDAQTHMEETSP	2.0	2007	C2	7
4WMS	MCL1	H. sapiens	149	DEALKDAQTGSELYRQ	1.9	2014	P21212	60
4WMTa	MCL1	H. sapiens	149	DEALKDAQTGSELYRQ	2.4	2014	P21212	60
4WMUa	MCL1	H. sapiens	149	DEALKDAQTGSELYRQ	1.6	2014	P21212	60
4WMVb	MCL1	H. sapiens	149	DEALKDAQTGSELYRQ	2.4	2014	P21212	60
4WMWb	MCL1	H. sapiens	149	DEALKDAQTGSELYRQ	1.9	2014	P21212	60
4WMXb	MCL1	H. sapiens	149	DEALKDAQTGSELYRQ	2.0	2014	P21212	60
4WGI	MCL1	H. sapiens	149	DEALKDAQTGSELYRQ	1.9	2014	P21212	59
3CSG	Monobody	NA	90	DEALKDAQTRITKGSSVPTNL	1.8	2008	P41	29
2OBG	Monobody	NA	90	DEALKDAQTRITKGSSVPTNL	2.3	2006	P41	25
3CSB	Monobody	NA	91	DEALKDAQTRITKGSSGSSVPTNL	2.0	2008	P41212	29
3WAI	Oligosaccharyltransferase C‐domain	A. fulgidus	368	DEALKDAQTNKQWYD	1.9	2013	C2	44
2VGQ	IPS‐1 CARD domain	H. sapiens	94	DEALKDAQTNSAMAFA	2.1	2007	P41212	26
3Q25	Alpha‐synuclein fragment	H. sapiens	19	DEALKDAQTNSSSMDVFM	1.9	2010	P43212	43
3Q26	Alpha‐synuclein fragment	H. sapiens	33	DEALKDAQTNSSSKAKEG	1.5	2010	P212121	43
3Q27	Alpha‐synuclein fragment	H. sapiens	26	DEALKDAQTNSSSKTKEG	1.3	2010	P21	43
3Q28	Alpha‐synuclein fragment	H. sapiens	22	DEALKDAQTNSSSKTKEQ	1.6	2010	P21	43
3Q29	Alpha‐synuclein fragment	H. sapiens	19	DEALKDAQTNSSSKTKEQ	2.3	2010	P212121	43
2ZXT	Tim40 core domain	S. cerevisiae	95	DEALKDAQTNSSSVPGRG	3.0	2009	P212121	27
1MG1	Gp21 ectodomain	HTLV	84	DAALAAAQTNAAAMSLAS	2.5	1999	H3	21
4OZQ	KIF14	M. musculus	349	DAALAAAQTNAAAENSQV	2.7	2014	P1	18
2XZ3	BLV TM hairpin	B. taurus	92	DAALAAAQTNAAALSHQR	1.9	2011	H3	8
4H1G	Kar3 motor domain	C. albicans	344	DAALAAAQTNAAALKGNI	2.1	2012	P21	48
3G7V	Amylin	H. sapiens	37	DAALAAAQTNAAAKCNTA	1.9	2011	P21	34
3G7W	Amylin	H. sapiens	22	DAALAAAQTNAAAKCNTA	1.7	2009	P41212	34
2NVU	NEDD8 activating enzyme E1	H. sapiens	429	DAALAAAQTNAAADWEGR	2.8	2006	P3221	24
3OB4	Ara h 2	A. duranensis	130	DAALAAAQTNAAARRCQS	2.7	2010	C2	–
4EGC	Six1	H. sapiens	189	DAALAAAQTNAAAMSMLP	2.0	2013	P21212	12
3O3U	RAGE	H. sapiens	210	DAALAAAQTNAAAASAQN	1.5	2010	P21212	41
4KYC	Polymerase foot domain	Menangle virus	50	DAALAAAQTNAAASGYKM	1.9	2013	P43212	15
4KYD	Polymerase foot domain	HPIV4b	50	DAALAAAQTNAAADALKI	2.2	2013	H3	15
4KYE	Polymerase foot domain	HPIV4b	50	DAALAAAQTNAAADALKI	2.6	2013	C2	15
4QVH	Phosphopantetheinyl transferase PptT	M. tuberculosis	227	DAALAAAQTNAAAMTVGT	1.7	2014	P212121	56
3PY7	E6	BPV1	152	DAALAAAQTNAAAMDDLD	2.3	2010	C2221	42
3F5F	2‐O‐Sulfotransferase	G. gallus	288	DAALAAAQTNAAADEEDD	2.6	2008	P63	33
4NDZ	2‐O‐Sulfotransferase	G. gallus	288	DAALAAAQTNAAADEEDD	3.4	2013	P212121	–
4KV3	EccD1 ubiquitin‐like domain	M. tuberculosis	90	DAALAAAQTNAAAMATTR	2.2	2013	P65	–
4EDQ	Myosin‐binding protein C	M. musculus	107	DAALAAAQTNAAADDPIG	1.6	2012	P1	–
3DM0	Rack1	A. thaliana	324	DAALAAAQTNAAAGLVLK	2.4	2008	P21	31
3H4Z	Der p 7	D. pteronyssinus	190	DAALAAAQTNAAADPIHY	2.3	2009	C2	9
4EXK	STM14_2015	S. enterica	103	DAALAAAQTNAAANTPSP	1.3	2012	P3221	—
4BL8	SUFU	H. sapiens	460	DAALAAAQTNAAAPGLHA	3.0	2013	P212121	46
4BL9	SUFU	H. sapiens	385	DAALAAAQTNAAAPGLHA	2.8	2013	P1	46
4BLA	SUFU	H. sapiens	385	DAALAAAQTNAAAPGLHA	3.5	2013	P21212	46
4BLB	SUFU	H. sapiens	382	DAALAAAQTNAAAPGLHA	2.8	2013	P21	46
4BLD	SUFU	H. sapiens	382	DAALAAAQTNAAAPGLHA	2.8	2013	P21	46
4PE2	PilA1 CD160	C. difficile	149	DAALAAAQTNAAASNINK	1.7	2014	I121	17
4TSM	PilA1 CD160	C. difficile	149	DAALAAAQTNAAASNINK	1.9	2014	C2	17
4OGM	PilA1 CD160	C. difficile	149	DAALAAAQTNAAASNINK	2.2	2014	P212121	17
3EF7	ZP3 ZPN domain	M. musculus	110	DAALAAAQTNAAAVKVEC	3.1	2008	P21212	30
3D4G	ZP3 ZPN domain	M. musculus	110	DAALAAAQTNAAAVKVEC	2.3	2008	P1	30
3D4C	ZP3 ZPN domain	M. musculus	110	DAALAAAQTNAAAVKVEC	2.9	2008	I222	30
4KEG	SPLUNC1	H. sapiens	196	DEALAAAQTNAAASPTGL	2.5	2013	P3221	–
4N4X	SPLUNC1	H. sapiens	214	DEALAAAQTNAAASPTGL	2.5	2013	P3221	–
4O2X	ClpS	P. falciparum	136	DEALAAAQTNAAANLEKI	2.7	2013	P61	–
4IKM	CARD8 CARD domain	H. sapiens	92	DAALAAAQTNAVDAAFVK	2.5	2012	P6522	50
4IFP	NLRP1 CARD domain	H. sapiens	84	DAALAAAQTNAVDLHFVD	2.0	2012	C2	49
3VD8	AIM2 PYD domain	H. sapiens	93	DAALAAAQTNAVDMESKY	2.1	2012	P212121	51
4WJV	Nsa2	S. cerevisiae	22	DEALAAAQTNAAAAMDTDG	3.2	2014	C2	19
1MH3	MATa1 homeodomain	S. cerevisiae	50	DAALAAAQTAAAAAISPQA	2.1	2002	P43212	22
1MH4	MATa1 homeodomain	S. cerevisiae	50	DAALAAAQTAAAAAISPQA	2.3	2002	P212121	22
4JBZ	Mcm10 coiled coil	X. laevis	32	DAALAAAQTNAAAMGVCQEK	2.4	2013	C2	14
3MP1	Sgf29 Tudor domain	S. cerevisiae	149	DEALAAAQTNAAAEFGSSYW	2.6	2010	P212121	38
3MP6	Sgf29 Tudor domain	S. cerevisiae	149	DEALAAAQTNAAAEFGSSYW	1.5	2010	P212121	38
3MP8	Sgf29 Tudor domain	S. cerevisiae	149	DEALAAAQTNAAAEFGSSYW	1.9	2010	P212121	38
1HSJ	SarR	S. aureus	115	DEALAAAQTNAAAEFMSKIN	2.3	2000	P1	2
1R6Z	Argonaute 2 PAZ domain	D. melanogaster	116	DEALKDAQTNAAAEFVDISH	2.8	2003	P65	4
4RWF	CALCLR:RAMP2	H. sapiens	217	DEALKDAQTNAAAEFGGTVK	1.8	2014	P212121	58
4RWG	CALCLR:RAMP1	H. sapiens	219	DEALKDAQTNAAAEFTTACQ	2.4	2014	C2	58
4MY2	Norrin	H. sapiens	103	DEALKDAQTNAAAEFIMDSD	2.4	2013	P21212	53
3C4M	PTH1R extracellular domain	H. sapiens	159	DEALKDAQTNAAAEFDDVMT	1.9	2008	P21	28
3L2J	PTH1R extracellular domain	H. sapiens	159	DEALKDAQTNAAAEFDDVMT	3.2	2009	C2221	37
4LOG	NR2E3 ligand binding domain	H. sapiens	194	DEALKDAQTNAAAEFLDSIH	2.7	2013	P21212	52
3N93	CFR2alpha extracellular domain	H. sapiens	102	DEALKDAQTNAAAEFAALLH	2.5	2010	C2221	–
3N94	PAC1R extracellular domain	H. sapiens	95	DEALKDAQTNAAAEFAIFKK	1.8	2010	P212121	40
3N95c	CRFR2ɑ extracellular domain	H. sapiens	102	DEALKDAQTNAAAEFAALLH	2.7	2010	P21	–
3N96c	CRFR2ɑ extracellular domain	H. sapiens	102	DEALKDAQTNAAAEFAALLH	2.7	2010	P21	–
3EHS	CRFR1 extracellular domain	H. sapiens	96	DEALKDAQTNAAAEFSLQDQ	2.8	2008	P41212	32
3EHT	CRFR1 extracellular domain	H. sapiens	96	DEALKDAQTNAAAEFSLQDQ	3.4	2008	P41212	32
3EHU	CRFR1 extracellular domain	H. sapiens	96	DEALKDAQTNAAAEFSLQDQ	2.8	2013	P1	32
4NUF	SHP	M. musculus	206	DEALKDAQTNAAAEFPHRTC	2.8	2013	P212121	54
4XAI	TLX ligand‐binding domain	T. castaneum	201	DEALKDAQTNAAAEFPSAIC	2.6	2014	P212121	61
4XAJ	TLX ligand‐binding domain	H. sapiens	202	DEALKDAQTNAAAEFTESVC	3.6	2014	P212121	61
3H3G	PTH1R extracellular doomain	H. sapiens	165	DEALKDAQTNAAAEFDDVMT	1.9	2009	P41212	35

NA, not applicable; RES., resolution (Å); YR., year deposited in PDB; S.G., space group; REF, reference.

^aIdentical MBP fusion proteins co‐crystallized with different ligands.

^bIdentical MBP fusion proteins soaked with different ligands.

Identical MBP fusion proteins co‐crystallized with different urocortin 2 peptides.

Why are there so many more structures of MBP fusion proteins in the PDB than of any other variety of fusion protein? In some cases MBP fusion protein structures simply have been presented as faits accomplis, with no backstory whatsoever.2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 In most instances, however, at least a brief description of the difficulties encountered with expression, solubility, stability, or crystallization of the target protein (or a combination of these factors) is provided as a rationale for crystallizing the MBP fusion protein.21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61 Remarkably, in only two cases is any mention made of attempts to crystallize proteins using multiple fusion partners. In one instance (1MG1), a GST fusion protein was also constructed, but it proved to be insoluble.21 In the other case, several different N‐terminal fusion partners were tested as crystallization chaperones for the MCL1 protein.60 The lysozyme fusion protein was insoluble. Although the corresponding SUMO, TRX, and MBP fusion proteins were soluble, only the latter yielded crystals (4WMS). It seems likely that the frequent choice of MBP as a fusion partner stems from a combination of its reputation for exerting a favorable impact on the yield, solubility, and stability of its fusion partners62, 63, 64 and its historical precedent as a successful crystallization tag.21, 65

Origin and Diversity of Inter‐Domain Linkers

The list of MBP fusion protein structures in Table 1 is organized according to the nature of their inter‐domain linkers. The linker residues (not part of either protein) are colored red. The first 11 entries are fusion proteins with relatively long linkers. The poly‐asparagine motif originates from the pMal‐C2 MBP fusion vector and its derivatives,66 which are marketed by New England Biolabs. If there is a logical explanation underpinning the choice of poly‐asparagine for a linker, it is by now forgotten and buried deeply in the historical record. The linker in 4O4B exhibits the signature sequence of Gateway recombinational cloning. The origin of the linker sequences in 1TOK, 1NMU, and 3A3C is uncertain. The existence of these 11 fusion protein structures appears to contradict the dogma that conformational flexibility between domains of a fusion protein is inimical to crystallization. That nearly all of the long linkers include a designed site for endoproteolysis (IEGR for Factor Xa or ENLYFQG for TEV protease) suggests that crystallization of these fusion proteins was undertaken as a fallback strategy (perhaps because proteolytic removal of the tag failed) rather than as the primary objective. Not surprisingly, there is no interpretable electron density for most of the linker residues in these structures.

The next six entries in Table 1 are fusion proteins with no inter‐domain linkers, which would be expected to have the greatest degree of rigidity. A slash mark (/) denotes the boundary between MBP and its fusion partners in these cases. There are a variety of ways to create translational fusions with seamless junctions (i.e., without residual cloning artifacts), the most common being overlap extension PCR.67 The C terminus of wild‐type MBP ends with the sequence DAQTRITK,68 which adopts an α‐helical conformation. The amino acid sequences reveal that one or more residues were deleted from the C terminus of MBP during the construction of these fusion proteins, although no rationale was provided for this in the accompanying publications.

The next group of linkers consists of the short sequences N, NS, HM, GS, GSS, AMD, GSSGSS, or NSSS. The latter linker is derived from the NEB vector pMal‐C2, wherein the three consecutive serine residues correspond to a unique SacI restriction site in the plasmid DNA that can be used for cloning.66 Hence, fusion protein expression vectors with the NSSS linker were constructed by ligating a passenger protein PCR product with an in‐frame SacI site at its proximal end. (“Passenger protein” is a term used to describe an open reading frame that is fused to the C terminus of MBP). In the case of the NS linker, the “sticky ends” generated by SacI were removed prior to ligation with a blunt‐ended PCR product.69 It is unclear what purpose is served by the asparagine residue that precedes the three serines; arginine occupies this position in wild‐type MBP. Yet, for whatever reason, the vast majority of the linker sequences in Table 1 include this amino acid substitution. Exceptions include the monobody fusion proteins with GSS (3CSG and 2OBG) or GSSGSS (3CSB) linkers.29

By far the most common inter‐domain linker consists of the four residues NAAA. There are 36 fusion protein structures with this linker sequence along with three amino acid substitutions near the C terminus of MBP (colored blue in Table 1). The three consecutive alanine residues coincide with a unique NotI restriction site in the plasmid DNA (Fig. 2). NotI recognizes the octanucleotide sequence GCGGCCGC. This enables passenger proteins to be joined in frame to the C terminus of MBP using the NotI site, which is added to the proximal end of the insert by PCR. The origin of the three point mutations near the C terminus of MBP can be traced back to the very first MBP fusion protein structure to have been determined (1MG1), the HTLV gp21 ectodomain fused to MBP.21 It was anticipated that the gp21 moiety would be a homotrimer, thus three of the charged residues near the C‐terminus of MBP were changed to alanine in a preemptive effort to avoid electrostatic repulsion. Although gp21 is indeed trimeric, the mutated residues were found not to be located in close proximity to one another in the crystal structure. It seems that this early vector design was subsequently disseminated widely in the structural biology community. There are a few variations on this theme. Several fusion proteins with the NAAA linker lack the proximal amino acid substitution in MBP. In 1MH3 and 1MH4, two of the earliest MBP fusion proteins to be crystallized, five consecutive alanine residues were utilized with a PstI site imbedded in them for cloning.22 In the 22 fusion proteins with the linker sequence NAAAEF, the EF sequence corresponds to an EcoRI site (GAA TTC) that was used for cloning (H. E. Xu, Personal Communication), although this was is not mentioned in the accompanying publications.28, 32, 35, 37, 40 Similarly, in the three fusion proteins with the linker sequence NAVD, the VD dipeptide probably corresponds to a SalI restriction site (GTC GAC), but this could not be verified from the published information.49, 50, 51

Figure 2.

Schematic diagram of a popular MBP fusion junction for crystallographic applications. See text for discussion.

Of course it would be useful to know if there is a particular linker or linkers that are most likely to yield useful crystals. Unfortunately, however, a definitive answer to this question is lacking because negative results are rarely reported. Nevertheless, the preponderance of the NAAA linker among the MBP fusion proteins that have been crystallized suggests that this would be a good choice.

Purification of Fusion Proteins

At this time eleven of the MBP fusion protein structures in the PDB have no accompanying publications (Table 1), and therefore no information is available about how these proteins were purified. The majority of the other 91 fusion proteins were purified first by amylose affinity chromatography and then by gel filtration. Consequently, the structures of most MBP fusion proteins in the PDB contain a ligand in the maltose‐binding pocket. In a small number of cases an additional purification step, usually ion exchange chromatography, was also employed.

MBP undergoes a substantial conformational change upon binding to maltose,70 and the two distinct conformations of MBP could give rise to different crystal contacts. For this reason, some fusion proteins were engineered to also include a hexahistidine tag attached either to the N terminus of MBP or the C terminus of the passenger protein. This enabled these fusion proteins to be partially purified by immobilized metal affinity chromatography rather than amylose affinity chromatography (with maltose), allowing crystallization screens to be performed both in the presence and absence of maltose if desired. In at least one case (3WAI), this strategy proved to be crucial, as only the ligand‐free form of the fusion protein could be crystallized.44 In another case, different crystal forms were obtained in the presence and absence of ligand.46 Curiously, however, even when MBP fusion proteins included a polyhistidine tag joined either to the N terminus of MBP (4O4B, 4JKM, 3MQ9, 4B3N, 2OK2, 3CSG, 2VGQ, 4WJV) or to the C‐terminus of the passenger protein (4IRL, 4BL8, 4BL9, 4BLA, 4BLB, 4BLD, 4PE2, 4TSM, 4OGM, 3EF7, 3D4G, 3D4C, 4O2X, 4IKM, 4IFP, 3VD8, 3C4M, 3L2J, 4LOG, 4RWF, 4RWG, 3N93, 3N94, 3N95, 3N96, 3EHS, 3EHT, 3EHU, 3H3G), most of the time amylose affinity chromatography was employed during purification or maltose was added to the protein prior to crystallization anyway. The use of both affinity techniques might facilitate the removal of truncated fusion proteins if the polyhistidine tag is joined to the C‐terminus of the passenger protein because then it would be flanked by N‐ and C‐terminal tags, creating an “affinity sandwich.” It is more difficult to understand why maltose was added to His‐MBP fusion proteins prior to screening for crystals.

Although, as discussed above, crystallization of an MBP fusion protein is generally undertaken as a fallback strategy after conventional methods have failed, Dr. H. Eric Xu of the Van Andel Research Institute has adopted a fusion‐based approach as a primary method for the production of class B G protein‐coupled receptor (GPCR) ligand‐binding domains.28 These (normally) extracellular domains, which often contain disulfide bonds, are produced in the cytosol of Escherichia coli as MBP fusion proteins. MBP maintains its fusion partners in a soluble state62, 64 while co‐expression of the bacterial protein disulfide isomerase DsbC in a trxB/gor strain71 aids in the oxidation and reshuffling of disulfide bonds. In some cases, additional disulfide reshuffling is carried out with partially purified fusion proteins in vitro. The MBP fusion proteins also have C‐terminal polyhistidine tags. Therefore, successive affinity chromatography on nickel and amylose columns ensures that only full‐length fusion proteins are purified. Residual aggregates are removed by gel filtration chromatography. The last 15 structures listed in Table 1 were obtained in this manner.

Phasing

In almost all cases, the initial phases used to solve the structures of MBP fusion proteins were obtained by molecular replacement, using MBP as a search model. The ability to obtain phases in this manner was recognized early on as a potential advantage of using MBP as a crystallization tag.1 At least 10 different MBP structures from the PDB, both ligand‐free and ligand‐bound, have been used as search models for molecular replacement. This approach has worked well because, at 42 kDa, MBP usually accounts for a substantial fraction of the X‐ray scattering atoms in the crystal lattice. In some cases, models of the passenger protein were used in addition to or instead of MBP for phasing. One of the fusion protein structures (4EGC) was phased de novo by anomalous scattering of selenomethionine residues.12 Another as yet unpublished structure (4IKM) contained iodotyrosine but evidently was solved by molecular replacement anyway.

One might expect that as the protein to which MBP is fused becomes larger and larger, the application of molecular replacement would become more difficult. Indeed, this appears to be the case. The structure of 4JKM, the largest MBP fusion protein to be crystallized, was solved by molecular replacement using a homolog of the 599 residue β‐glucuronidase fusion partner as a search model rather than MBP.20 The structure of MBP‐NEDD8 E1, which was crystallized as a multiprotein complex (2NVU), was also phased by molecular replacement without the assistance of MBP.24 The structure of 3WAI, on the other hand, which has a fusion partner of approximately the same size as MBP, was solved by molecular replacement with MBP, although the phases were subsequently improved by anomalous scattering from selenomethionine residues.44 Other MBP fusion protein structures with very large fusion partners (4H1G, 3DM0,4BL8, 4BL9, 4BLA, 4BLB, 4BLD) were phased by molecular replacement using a combination of MBP and a homolog of the fusion partner as search models.31, 46, 48

General Characteristics of MBP Fusion Protein Structures

Passenger proteins range in length from 19 to 599 residues (Table 1). About 71% of the passenger proteins are fewer than 150 residues in length, with the greatest number being between 101 and 150 amino acids long [Fig. 3(a)]. It is not known whether large passenger proteins crystallize less readily than smaller ones when fused to MBP or if more attempts have been made to crystallize small fusion proteins than large ones. Some of the shortest passengers are fragments of larger proteins (<60 residues) that may be too small to adopt stably folded conformations.

Figure 3.

General characteristics of MBP fusion proteins. (a) lengths of passenger proteins in intervals of amino acid residues. (b) Resolution of MBP fusion protein structures in Å (black bars, left ordinate scale) versus the sum total of all unique structures in the PDB (gray squares, right ordinate scale). (c) percentage of MBP fusion proteins crystallized in each of 16 different space groups (black) and percentage of all unique proteins in the PDB that have been crystallized in the same space groups (gray).

The resolution of MBP fusion protein structures resembles a normal distribution between 1.3 and 3.5 Å with an average slightly higher than 2 Å [Fig. 3(b)]. Sixty‐seven percent of the structures are 2.5 Å resolution or better. The average resolution of all structures in the PDB is slightly higher, but it should be noted that the average molecular weight of the MBP fusion proteins is greater than the average molecular weight of all structures in the PDB. MBP fusion proteins have been crystallized in 16 different space groups. Their relative frequency parallels that of the PDB as a whole, with P2₁2₁2₁, P2₁2₁2, P2₁, and C2 being the most common [Fig. 3(c)]. The average solvent content of the 102 MBP fusion protein crystals is 54% compared with an average of 50% for all protein crystals in the PDB. The most remarkable thing about these structures is how unremarkable they are: in virtually all respects, they are rather typical of other structures in the PDB.

While the vast majority of MBP fusion protein structures are single polypeptides, a handful of them have been crystallized as multiprotein complexes or in complex with oligopeptides. Examples of the latter type are 4WJV, 4RWF, 4RWQG, 3N93, 3N95, 3N96, 3EHT, 3EHU, 4NUF, 4XAI, 4XAJ, and 3H3G. In 4EGC, MBP fused to human Six1 was co‐crystallized with Eya domain of Eya2.12 The most spectacular example of a multiprotein complex containing an MBP fusion protein to have been crystallized thus far is the ubiquitin‐like protein NEDD8 in complex with its heterodimeric E1 ligase.24 In this complex, the catalytic subunit of the E1 ligase is fused to MBP. There is also one structure of an MBP fusion protein in complex with an RNA molecule (1T0K): Saccharomyces cerevisiae ribosomal protein L30 with a fragment of rRNA.5

Surface Entropy Reduction Mutants

A protein that fails to crystallize can sometimes be induced to do so by replacing clusters of long, flexible amino acid side chains on its surface with alanine residues. This general technique, termed surface entropy reduction mutagenesis (SERM), was originally described by Derewenda and his colleagues.72 The Pedersen laboratory was the first to combine the use of MBP as a crystallization chaperone with the SERM technique.65 Nineteen of the MBP fusion protein structures deposited in the PDB include surface entropy reduction mutations in MBP (Table 2). The amino acid substitutions are D83A/K84A, E173A/N174A, K240A, or a combination of these. In 14 of these 19 structures, one or more of the mutated residues participates in crystal contacts (being closer than 4.5 Å apart73), as has frequently been observed in other instances of SERM.74 A detailed analysis of the crystal contacts mediated by the mutated residues in these structures failed to reveal any conserved interactions between MBP molecules (data not shown). Intuitively, and in the absence of any evidence to the contrary, it seems that the most promising strategy may be to utilize the MBP variant in which all of the surface entropy reduction mutations are present simultaneously. The introduction of additional mutations might further increase the probability of obtaining crystals.

Table 2.

Surface Entropy Reduction Mutations in MBP and Their Participation in Crystal Contacts

PDB code	D83A/K84A	E173A/N174A	K240A	Reference
3DMO	✓		✓ (+)	–
3H4Z	✓ (+)	✓ (+)	✓ (+)	9
4EXK	✓ (+)	✓ (+)	✓ (+)	–
4IKM	✓	✓	✓	49
4 IFP	✓ (+)	✓	✓ (+)	48
3VD8	✓*	✓*	✓	50
4IRL	✓ (+)	✓	✓ (+)	13
4JBZ	✓	✓ (+)	✓	14
3OB4	✓	✓	✓ (+)	–
4EGC		✓ (+)		12
4KYC		✓		15
3PY7	✓		✓	41
4QVH		✓ (+)		55
4WJV	✓		✓ (+)	19
4OGM	✓ (+)	✓	✓	17
4TSM	✓	✓ (+)	✓	17
4KYE	✓ (+)			15
4KYD	✓ (+)			15
4GIZ	✓		✓	41

^✓ Signifies the presence of the indicated surface entropy reduction mutations in MBP. (+) Indicates participation of the mutated residue(s) in one or more crystal contact (closer than 4.5 Å). *N174A is close to K84A in an adjacent molecule (7.3 Å). Although not close enough to constitute a crystal contact, the original side chains probably would have clashed in this packing arrangement.

Impact of MBP on the Structure of its Fusion Partners

Perhaps the greatest concern about the crystallization chaperone approach in general is that the chaperone, MBP in this case, may distort or otherwise interfere with the conformation of its fusion partner. In other words, there may be significant structural differences between the fused and unfused forms of a protein. In 24 cases, it was possible to compare the structures of unfused proteins solved by X‐ray crystallography or NMR with those of the corresponding MBP fusions to assess how frequently such differences occur. All pairs of protein structures that have been determined in both fused and unfused states were aligned using PDBeFold75 and PyMOL76 and the results are summarized in Table 3.

Table 3.

Structural Alignments of Fused and Unfused proteins

MBP fusion	Unfused	R.M.S.D. (Å)	Matches
4O4B	4O4C	0.74	233/239
3OAI	1NEU	0.94	113/115
4B3N	2LM3a	0.63	175/202b
2OK2	3ZLJ	0.93	31/31
3WAI	3WAJ	0.52	294/368b, c
2VGQ	4P4H	0.67	93/94
1NMU	1CK2a	1.65	100/104
3G7V	2KB8a	3.97d	25/37b
2NVU	3GZN	0.65	338/429e
3O3U	4LP4	1.05	209/210
4KV3	4KV2	0.55	90/90
4EDQ	3CX2	0.84	105/107
3H4Z	3UV1	0.95	187/190
4EXK	2LV4a	1.43	103/103
4BL8	4KM9	1.30	354/371b, f
4IFP	3KAT	0.71	82/84
3VD8	4O7Q	1.03	91/93
4KEG	4KGO	1.20	160/196b
1R6Z	1VYNa	1.71	112/116
4QVH	4QJK	0.68	224/227
4WMS	4WMR	1.17	145/149
4WMTg	4WMR	0.82	148/149
4WMUg	4WMR	0.92	148/149
4EXK	2LV4a	1.43	103/103
4N4X	4KGH	1.01	157/197b

^aNMR structure.

^bMissing electron density for segment(s) of the passenger protein.

^cConformational shift in the position of N‐terminal domains, but N‐terminal domains align very well.

^dBoth of these polypeptides adopt an α‐helical conformation, but the helices do not align well.

^eConformational shift in the position of C‐terminal domains, but C‐terminal domains align very well.

^fC‐termini do not align well.

^gThese fusion proteins are identical to 4MWS but were co‐crystallized with two different small molecules.

The average r.m.s.d. for C‐α atoms between the fused passenger proteins and their unfused counterparts is on the order of approximately 1 Å in most cases, indicating very close agreement between their structures. Minor differences between the conformations of fused and unfused proteins occur most frequently at the ends of the passenger proteins and less frequently in the loops between secondary structure elements. Not unexpectedly, larger r.m.s.d. values are usually observed when the unfused structure was determined by NMR (e.g., 1CK2, 2KB8, 2LV4, 1VYN), but in none of these cases are the structures radically different.

Large conformational shifts occur in only two pairs of fused and unfused structures. The alignment between 3WAI and 3WAJ returns an average r.m.s.d. for C‐α atoms of just 0.52 Å over 294 of 368 residues. The non‐matching residues correspond to the N‐terminal about 70 residues of the polypeptide that is fused to MBP, which is only a fragment of the full‐length Archaeoglubus fulgidus AglB enzyme. On the other hand, in the structure of unfused, full‐length AglB (3WAJ) these 70 residues are preceded by a large α‐helical N‐terminal domain and they are sandwiched between it and the remainder of the protein, effectively locking them into the observed conformation.77 The N‐terminal α‐helical domain is absent from the truncated AglB that is fused to MBP so it is unable to constrain the movement of the 70 N‐terminal residues of this AglB fragment. Additionally, in the structure of the MBP fusion protein, the C‐terminal α‐helix of MBP forms a contiguous, extended helix with the N‐terminal segment of the AglB fragment, which may contribute to its differing orientation in the fusion protein. The conformational shift of this “mini domain” can be seen when the fragment of AglB from the MBP fusion protein structure is aligned with the corresponding residues from the structure of the full‐length protein [Fig. 4(a)]. However, the structures of the two mini domains align very well [Fig. 4(b)], indicating that although their relative positions are different in 3WAI and 3WAJ, their folds are unchanged. This is the only known case in which there is some evidence to suggest that MBP may have been at least partially responsible for a structural distortion (rearrangement) of its fusion partner.

Figure 4.

Conformational shifts observed in fused and unfused structures. (a) structural alignment of 3WAI (magenta) and 3WAJ (blue). (b) structural alignment of N‐terminal “mini domains” of AglB in 3WAI (magenta) and 3WAJ (blue). (c) structural alignment of MBP‐UBA3 from 2NVU (magenta) with unfused UBA3 from 3GZN (blue). (d,) structural alignment of the ubiquitin fold domains from fused (magenta) and unfused (blue) UBA3.

The alignment between the relevant polypeptide chains in 2NVU and 3GZN yields an average r.m.s.d. of 0.65 Å for matching residues (338/429). This case is complicated by the fact that both the fused and unfused structures have been crystallized as multiprotein complexes. The complexes are composed of the ubiquitin‐like protein NEDD8 in complex with its heterodimeric E1 activating enzyme, which consists of a regulatory domain (APPBP1) and a catalytic domain (UBA3). In one complex (2NVU), UBA3 is fused to the C terminus of MBP.24 The non‐matching residues in the UBA3 alignment can be ascribed to a substantial conformational shift in the orientation of its C‐terminal domain (also known as the ubiquitin fold domain or UBF) [Fig. 4(c)], yet the fold of this domain is the same in the two molecules [Fig. 4(d)]. An additional protein, the E2 enzyme Ubc12, is present in the complex with unfused UBA3 (3GZN).77 According to Huang et al.,24 the conformational shift of the UBA3 UBF domain is required to create the binding surface for Ubc12. Moreover, in the 2NVU structure, all crystal contacts between MBP are with regions of APPBP1, UBA3, and NEDD8 that have identical conformations in previous structures that do not include MBP.24, 78 Therefore, this structural rearrangement is not caused by MBP.

In summary, based on the examination of 24 pairs of fused and unfused structures, it seems that no major structural derangements have occurred entirely as a consequence of being fused to MBP. This does not preclude their existence, however. It may be that MBP‐induced structural distortions do occur in some cases but fail at the expression, purification, or crystallization stages and so they are not observed.

One might also wonder if the structure of MBP is altered by its fusion partners. When structural alignments of the MBP domains from all 105 fusion protein structures were performed with high resolution structures of unfused MBP (either 1ANF for the ligand‐bound conformation or 1LLF for the unliganded protein), no significant structural deviations were detected (data not shown), except for 4JKM wherein the N‐terminal residues of MBP may have been traced incorrectly due to poor electron density in this region.

Future Directions

The growing number of MBP fusion protein structures in the PDB is evidence that the crystallization chaperone strategy is being utilized more frequently, and we can expect the number of such structures will continue to increase. As shown above, there is little evidence that MBP alters the structure of its fusion partners. MBP appears to be a generally effective crystallization chaperone, but other proteins may be equally or even more effective. Thus far, however, very little effort has been made to identify alternatives to MBP. Additionally, the repertoire of tools for chaperone‐assisted crystallization might be further expanded by employing proteins that have been selected to bind tightly to MBP, such as monobodies,25, 29 Fab fragments79, 80 and Designed Ankyrin Repeat Proteins (DARPins)81 for co‐crystallization with MBP fusion proteins. This approach may increase the probability of obtaining crystals without the need to express and purify more than one MBP fusion protein.

Methods

Fusion protein structures were identified by using the Basic Local Alignment Search Tool (BLAST)82 server to query the PDB with the amino acid sequences of E. coli MBP,68 E. coli TRX,83 and Schistosoma japonicum GST.84 The matches were then inspected manually to identify the fusion proteins. Crystal contacts involving surface entropy reduction mutations in MBP were identified by manual inspection using the “measure” tool in PyMOL.76 Structural alignments were performed with the PDBeFold server75 and PyMOL76 after removing irrelevant atoms from the PDB files. Data from the Protein Data Bank (PDB) that were used in this analysis were extracted in August 2015.