Pc

AC

a

ARRAA

KABDpl

1

bmtmSbRlmicpTp

h0

Journal of Chromatography A, 1343 (2014) 119–127

Contents lists available at ScienceDirect

Journal of Chromatography A

j o ur na l ho me page: www.elsev ier .com/ locate /chroma

urification of pre-miR-29 by a new O-phospho-l-tyrosine affinityhromatographic strategy optimized using design of experiments

driana Afonso, Patrícia Pereira, João A. Queiroz, Ângela Sousa, Fani Sousa ∗

ICS-UBI—Health Sciences Research Centre, University of Beira Interior, Avenida Infante D. Henrique, 6200-506 Covilhã, Portugal

r t i c l e i n f o

rticle history:eceived 19 January 2014eceived in revised form 14 March 2014ccepted 27 March 2014vailable online 4 April 2014

eywords:ffinity chromatographyox–Behnken designesign of experimentsre-miR-29-Tyrosine

a b s t r a c t

MicroRNAs are the most studied small non-coding RNA molecules that are involved in post-transcriptional regulation of target genes. Their role in Alzheimer’s disease is being studied and exploredin order to develop a new therapeutic strategy based on specific gene silencing. This disease is charac-terized by protein deposits, mainly deposits of extracellular A� plaques, produced upon endoproteolyticcleavage of APP by ß-site APP-cleaving enzyme 1 (BACE1). Recent studies have shown that particularlymiR-29 cluster can be involved in the decrease of A� plaques production, by acting on BACE1 expressionsilencing. In order to use this microRNA as potential therapeutic it is essential to guarantee its purity,stability and integrity. Hence, the main purpose of this study was the development of a new affinity chro-matographic strategy by using an O-phospho-l-tyrosine matrix and applying Box–Behnken design (BBD)to obtain pre-miR-29 with high purity degree and yield, envisioning its application in gene therapy. Thus,after process optimization the best results were achieved with a decreasing ammonium sulfate gradientin 10 mM Tris buffer, pH 8 (1.6 M (NH4)2SO4, 1.11 M (NH4)2SO4 and 0 M (NH4)2SO4), at 16 ◦C. These exper-

imental conditions allowed the recovery of pre-miR-29 with 52% of purity and 71% of recovery yield. TheO-phospho-l-tyrosine matrix was initially chosen to mimic the natural interactions that occur inside thecell, and in fact it was proved a satisfactory selectivity for pre-miR-29. Also the innovative application ofBBD for this strategy was efficient (R2 = 0.98 for % relative recovery and R2 = 0.93 for % relative purity) andessential to achieve best purification results in short time, saving lab resources.

© 2014 Elsevier B.V. All rights reserved.

. Introduction

MicroRNAs (miRNAs) are small non-coding RNAs, with 15–21ases, that bind partially to complementary sequences in targetessenger RNA (mRNA), silencing them, by preventing transla-

ion or by inducing its degradation [1,2]. Due to their functions,iRNAs represent a potential and interesting therapeutic target.

everal pathological features of Alzheimer’s disease (AD) haveeen recognized and linked to a deregulation of many miRNAs [3].ecently, Hebert and co-workers [4] proved that the decreased

evels of microRNA-29a/b-1 cluster were connected to abnor-ally high levels of BACE1 protein expression [4]. This enzyme

s directly associated to the production of Aß peptides, once itleaves the amyloid precursor protein (APP), producing the A�-

eptides, which aggregate and originate the �-amyloid plaques [5].his statement raised the chance of using pre-miR-29 as a thera-eutic target. Therefore, it is essential to guarantee the suitable

∗ Corresponding author. Tel.: +351 275 329 074; fax: +351 275 329 099.E-mail address: fani.sousa@fcsaude.ubi.pt (F. Sousa).

ttp://dx.doi.org/10.1016/j.chroma.2014.03.071021-9673/© 2014 Elsevier B.V. All rights reserved.

preparation of pre-miR-29, preserving its purity, stability andintegrity. Currently, several methods have been already describedfor RNA isolation and/or purification, namely phenol-based extrac-tions, silica matrix or glass fiber filter-based binding methods [6].In addition, elution of RNAs from denaturing polyacrylamide gelelectrophoresis, magnetic oligo(dT) beads for the purification ofpoly(A)+ RNA, anion exchange HPLC, size exclusion chromatog-raphy [7] and affinity chromatography using RNA aptamers orstreptavidin matrices associated with biotinylated RNAs were alsostudied. In general, these methods are mostly time consuming,expensive and tedious and the usage of synthetic RNAs also repre-sents a limitation [7]. More specifically, RNAs biotinylation errorscan occur, and in this case RNAs are permanently linked to biotinmolecule and will always have impurities resulting from reaction[8]. Some tagged RNAs do not cleave even after prolonged incuba-tion, which can significantly degrade the molecule and oligomericRNA species formed during transcription cannot be separated from

the monomeric species [9]. However, amino-acid affinity chro-matographic methods have been studied as a good alternative tothe classic methods, which are based on the natural interactionsthat occur between amino acids and nucleic-acids on a natural

1 atogr.

eloba

t[pmtutrtttategnswotawatot

2

2

sW(s(tLaP(0ag(u

2

pBtguewi

20 A. Afonso et al. / J. Chrom

nvironment [10]. The purification occurs due to the biospecificigands (amino acids) that recognize the biomolecules on the basisf a reversible interaction (based on electrostatic and hydropho-ic interactions, van der Waals forces and hydrogen bonding) thatllows the purification of the biomolecule in one single step [10].

Our research group have further explored this field, and thisechnology has been successfully applied to purify plasmid DNA10–12], some RNA species like 6S [13] and small RNAs [14–16]. Theresent work suggests the use of an O-phospho-l-tyrosine agaroseatrix to purify pre-miR-29, for potential application as therapeu-

ic agent on Alzheimer’s disease. This matrix has been previouslysed to purify proteins [17] and now the same matrix is proposedo purify a specific nucleic acid. This choice was based on miRNAecognition by the RNA-induced silencing complex (RISC) insidehe cell, through a conserved tyrosine residue [18,19], essential tohe right cleavage of mRNA. The conserved tyrosine is crucial tohe bound between RISC and miRNA [18] as its absence attenu-te the mRNA cleavage activity [18,20]. Consequently, the aim ofhis study was to explore these natural interactions, which werexpected to occur as equal between the miRNA and the chromato-raphic matrix. The optimization of chromatographic methods inucleic acids purification is often intricate and can be time con-uming. Usually, it is achieved by changing parameters one by onehich is commonly called “one factor/variable at a time” method-

logy or simply “one at a time” method [21]. In the present work,he Box–Behnken design was used as an innovative approach tochieve the optimization for the purification method. This designas used to maximize the efficiency of scientific experimentations

nd minimize waste and cost [22,23]. The application of this par-icular approach to this case allowed for an improved developmentf the chromatographic process diminishing the effort with respecto search for the best conditions.

. Materials and methods

.1. Materials

�-mercaptoethanol, sodium acetate, chloroform (Merck, Darm-tadt, Germany), isopropanol (Thermo Fisher Scientific Inc.,

altham, USA), guanidinium thiocyanate salt and sodium citrateSigma-Aldrich, St. Louis, USA) were used as reagents to cell lysis. Allolution were freshly prepared using 0.05% Diethyl pyrocarbonateDEPC) (Sigma-Aldrich, St. Louis, USA) treated water. For purifica-ion, an O-phospho-l-tyrosine agarose matrix (Sigma-Aldrich, St.ouis, USA) and different salts for buffers were used, such as, 2-mino-2-hydroxymethyl-propane-1,3-diol (Tris) (Nzytech, Lisbon,ortugal), ammonium sulfate ((NH4)2SO4) and sodium chlorideNaCl) (VWR, Pennsylvania, USA). All buffers were filtered through a.20 �m pore size membrane (Schleicher Schuell, Dassel, Germany)nd degassed ultrasonically. For cDNA evaluation, using an agaroseel electrophoresis, a DNA molecular weight marker was usedVivantis Technologies, Selangor DE, Malaysia). All the materialssed in the experiments were RNase-free.

.2. Small RNA production and isolation

The RNA used in this study was obtained from the marinehotosynthetic bacterium R. sulfidophilum DSM 1374 (BCCM/LMG,russels, Belgium) previously transformed with pBHSR1-RM vec-or [24], genetically modified with pre-miR-29 sequence. Bacterialrowth was carried out at 30 ◦C under dark-aerobic conditions,

sing Nutrient Broth medium (1 g/L beef extract; 2 g/L yeastxtract; 5 g/L peptone and 5 g/L sodium chloride) supplementedith 30 �g/mL kanamycin. Cell growth was evaluated by measur-

ng the optical density of the culture medium at a wavelength of

A 1343 (2014) 119–127

600 nm. The small RNAs samples were obtained using an adaptationof the acid guanidinium thiocyanate–phenol–chloroform methodbased on what has been described by Martins and coworkers [13].Briefly, the bacterial pellets were lysed and the nucleic acid frac-tion obtained was precipitated with isopropanol. After centrifugingfor 20 min at 16,000 × g (4 ◦C), the pellets were air-dried and thensolubilized in 1 mL of 0.05% DEPC-treated water. Finally, the RNAconcentration and purity was determined spectrofotometrically,using a Nanodrop spectrophotometer and by agarose gel elec-trophoresis.

2.3. Affinity chromatography

Affinity chromatography was performed in an ÄKTA Avant sys-tem with UNICORNTM 6.1 software (GE Healthcare Biosciences,Uppsala, Sweden) using the sRNAs samples obtained from R.sulfidophilum. A 20 mL column was packed with commercial O-phospho-l-tyrosine agarose gel. This support is characterized bythe manufacturer as a cross-linked 4% beaded agarose matrixwith a 1-atom spacer and an extent of labeling between 5 and15 �mol/mL. To test the main interactions occurring between RNAsand the matrix, 200 �g/mL of sample were injected on the col-umn through a 200 �L loop capilar. Electrostatic interactions wereexplored by using a stepwise gradient from 10 mM Tris buffer (pH8) to 1 M NaCl in 10 mM Tris buffer (pH 8). On the other hand,hydrophobic interactions were favored using high (NH4)2SO4 con-centration in 10 mM Tris buffer (pH 8) as equilibrium and bindingsolution and 10 mM Tris buffer (pH 8) to elute the bound RNAs. Inparticular, the column was equilibrated with 2 or 1.4 M (NH4)2SO4in 10 mM Tris buffer (pH 8), depending on the experiments per-formed, using a flow rate of 1 mL/min. After elution of unboundspecies the ionic strength of the buffer was decreased using a step-wise gradient in a range of 1.3 to 1 M of (NH4)2SO4 in 10 mM Trisbuffer (pH 8), according to the experiments performed. Finally,the column was washed, removing tightly bound RNA speciesby changing to 10 mM Tris buffer (pH 8). The absorbance of theeluate was continuously monitored at 260 nm. All experimentswere performed at controlled temperature (4 ◦C to 20 ◦C) usinga circulating water bath. Fractions were pooled according to thechromatograms obtained, and then concentrated and desalted withVivaspin concentrators (Sigma-Aldrich, St. Louis, USA). All sampleswere reserved (−80 ◦C) for further analysis.

2.4. Polyacrylamide gel electrophoresis

The integrity and identification of sRNAs recovered from thechromatographic experiments was checked by denaturating ureapolyacrylamide gel electrophoresis using the Amersham Bio-sciences system (GE Healthcare Biosciences, Uppsala, Sweden). AllsRNAs samples (20 �L) were denatured with 97.5% formamide, 0.3%bromofenol, 10 mM EDTA at pH 7.5 solution and denatured condi-tions were kept in the gel due to the presence of 8 M urea. Then,sRNAs were resolved into a 10% polyacrylamide gel and run at 120 Vfor 120 min with TBE buffer (0.84 M Tris base, 0.89 M boric acid and0.01 M EDTA, pH 8.3). After the run, gels were stained with ethidiumbromide (0.5 mg/mL).

2.5. Experimental design for optimization of pre-miR-29purification

The Box–Behnken design (BBD) was used to identify optimumconditions in the separation process. The studied factors were

the binding and elution conditions and column temperature. Thisselection was based on the most three critical factors that affectthe chromatographic process and were studied at three levels.To choose the range of each level it has been used some of the

A. Afonso et al. / J. Chromatogr. A 1343 (2014) 119–127 121

Table 1Factors and respective low and high values used for optimization of pre-miR-29 purification.

Code Factor Levels

Low Medium High−1 0 1

kteTwtlpotbfipaatsFa

2

cw7mreTs

2

cas

TBa

A Binding [(NH4)2SO4] in 10 mM Tris buffer (pH 8)

B Temperature (◦C)C Elution [(NH4)2SO4] in 10 mM Tris buffer (pH 8)

nowledge obtained from preliminary experiments using “one at aime method”. Each variable was coded by letters A, B and C, andach one had 3 levels which were −1, 0 and 1 as shown in Table 1.he design matrices were generated using UNICORNTM 6.1 soft-are, resulting on 15 experiments to be performed (Table 2) with

wo more replicates of the central point. Two responses were estab-ished for the determination of optimum method conditions: theercentage of relative recovery of pre-miR-29 (% relative recoveryr R1) and the percentage of relative purity of pre-miR-29 (% rela-ive purity or R2). The percentage of relative recovery was obtainedy measuring the intensity of the pre-miRNA band (where it is puri-ed) relatively to the sum of all pre-miR-29 bands obtained in alleaks. On the other hand, the percentage of relative purity wascquired measuring the intensity of the pre-miRNA band relative toll the other bands present within the pre-miRNA peak. To measurehe intensity of electrophoresis bands it was used the Phoretix 1Doftware (Nonlinear Dynamics, Ltd., Newcastle, United Kingdom).urthermore, the fractions of each peak were collected, desaltednd concentrated and cDNA was produced to perform a PCR.

.6. Model validation and purification process optimization

For the model validation three assays were performed as repli-ates of the optimum point. The optimum separation conditionsere identified introducing the data of DoE on the Design Expert

.0 software (State Ease Inc., Minneapolis, USA). The model wasanaged to predict the best purification, even compromising the

elative recovery. For that, the importance of these responses wasstablished as 2 for % relative recovery and 5 for % relative purity.he intensity of the bands was again evaluated with Phoretix 1Doftware.

.7. cDNA synthesis

cDNA was synthesized using the “RevertAidTM First StrandDNA Synthesis Kit” (Thermo Fisher Scientific Inc., Waltham, USA),ccording to manufacturer’s instructions. A total of 0.5 �g of RNAamples collected after the chromatographic purification process

able 2ox–Behnken design matrix along with the two responses for each run (R1—% rel-tive recovery and R2—% relative purity).

Run no. Binding (M) Elution (M) Temperature (◦C) R1 R2

1 2 1.15 4 66.23 12.822 1.4 1.15 4 6.93 14.203 2 1.15 20 96.12 7.264 1.4 1.15 20 4.90 15.855 2 1.3 12 69.83 13.736 1.4 1.3 12 0.20 0.007 2 1 12 96.12 7.268 1.4 1 12 11.39 21.339 1.7 1.3 4 36.17 8.2110 1.7 1.3 20 55.42 16.3811 1.7 1 4 76.86 16.3112 1.7 1 20 98.93 21.2313 1.7 1.15 12 78.27 21.2514 1.7 1.15 12 85.36 31.2115 1.7 1.15 12 77.70 24.16

2 1.7 1.44 12 201.3 1.15 1

with tyrosine–agarose column were used to initiate cDNA synthe-sis. Specific primers of gene (5′ GAC AGC GGT ATG ATC CCC CAA3′) were added to the RNA. At the end, cDNA was obtained andready to be used on polymerase chain reaction (PCR) amplificationto know if the pre-miRNA was eluted on a specific peak. First, a mixwas prepared with 0.125 U of Supreme DNA polymerase (NZYtech,Lisbon, Portugal), 50 mM of magnesium chloride (NZYtech, Lisbon,Portugal) and 150 nM of each primer. Specific primers (Forward–5′-GGA AGC TGG TTT CAT ATG GTG-3′ and Reverse–5′-CCC CCA AGAACA CTG ATT TC-3′) for cDNA witch design was achieved onRNA database, were used to amplify a fragment of 145 bp. Thepolymerase chain reaction was already optimized for the primersinvolved and the program followed was: 5 min at 95 ◦C for denatur-ation of cDNA, 35 cycles with three different steps − 95 ◦C for 30 s,63 ◦C for 30 s and 72 ◦C for 15 s – for cDNA amplification with 63 ◦Cas annealing temperature and 5 min at 72 ◦C for final elongation. Toconfirm the presence and purity of amplicons, PCR products wereanalyzed by 1% agarose gel electrophoresis (110 V for 30 min) inTAE buffer in DEPC-treated water, stained with 0.3 �L/mL green-safe (NZYtech, Lisbon, Portugal) and visualized under UV light in aViberLourmat system (ILC Lda, Lisbon, Portugal).

3. Results and discussion

Alzheimer’s disease represents a health, economical and publicburden, it being necessary to found a useful therapeutic agent toovercome the disease and all these problems. Recent studies havedemonstrated that miRNAs are involved on the control of this dis-ease [25] and that, particularly, the cluster miR-29 is diminishedon AD brains [4]. In order to use miRNA-based therapies, a pure,stable and active molecule should be obtained. In this work, an O-phospho-l-tyrosine matrix was used for the first time to purify thepre-miR-29 from a recombinant system. Exploiting the specific andnatural interactions between miRNA and the affinity matrix bringsseveral advantages, because it is not necessary to modify the recom-binant pre-miRNA in order to bind it to the matrix and it is not useda synthetic pre-miRNA, since this method allowed to purify thetarget molecule after its extraction from cells. To reach this objec-tive, an optimization of the chromatographic method should bedone. Thus, using a Box–Behnken design approach it was possibleto achieve the rapid pre-miRNA purification and test several con-ditions that interfere with the purification process, accessing theirinfluence on purification performance and optimizing it through amathematical model.

3.1. Pre-miR-29 and O-phospho-l-tyrosine agarose matrixinteractions

Amino-acid affinity chromatography is based on biologicalspecificity, which occurs between amino acids and nucleic acids.This type of selectivity is explained by the biological or chemicalstructure recognition, which favors the interaction [10]. Partic-

ularly, the use of an O-phospho-l-tyrosine (P-Tyr) brings thechance to explore the natural interaction that occurs betweenRISC complex, involving a conserved tyrosine residue and miRNAmolecules [18,20]. The interactions between the amino acid ligand

122 A. Afonso et al. / J. Chromatogr. A 1343 (2014) 119–127

Table 3ANOVA table for Box–Behnken design model.

Response Degrees of freedom Sum of squares Mean of square F-value P-value Lack of fit R2 Adj R2 Adeq Precision

277

ahbap

eRl2ratctscmbesgooit

3

omtmaasugRtw7lv

R

R

msdc

R1 9 17,503.4 1944.83

R2 9 751.13 83.46

nd the target molecule can be mostly based on electrostatic andydrophobic interactions, van der Waals forces and hydrogenonding. For elution, the interaction can be reversed by using

competitive ligand, or by changing the pH, ionic strength orolarity of the mobile phase [26].

In the present work, when electrostatic conditions were mainlyxploited by using 10 mM Tris buffer followed by 1 M NaCl, allNAs eluted on 10 mM Tris buffer step (data not shown), revea-

ing no interaction with the column. On the other hand, when M (NH4)2SO4 in 10 mM Tris buffer (pH 8) was used, the RNAsemained bound to the matrix, showing that hydrophobic inter-ctions were the main interactions occurring between RNAs andhe O-phospho-l-tyrosine ligands. Furthermore, by analysis of theDNA of each peak, it was confirmed that the pre-miRNA bound tohe column was only eluted after decreasing the conductivity (ionictrength) of the elution buffer (data not shown). These findingsan be explained by the presence of hydrophobic groups on P-Tyratrix, as the aromatic group, that will promote the hydropho-

ic interaction [27] with pre-miR-29. The interaction mechanism isxplained by the salt effect on the reduction of the water moleculesurrounding the target biomolecules, exposing their hydrophobicroups and making possible the recognition by the ligands. More-ver, the structure of pre-miRNA, a stem-loop or hairpin consistingf two long irregular double-stranded stem regions, which arenterrupted by a largely single-stranded internal loop, contributedo improve the hydrophobic interactions with the matrix.

.2. Design of experiments

The Box Behnken design (BBD) revealed to be a good tool toptimize the pre-miR-29 purification by using an affinity chro-atography method. Besides the fact this design extremely reduce

he number of experiments needed to achieve this kind of infor-ation, it does not contain combinations where all the factors are

t their higher or lower levels [28]. These conditions represented huge advantage for the present work because it was not neces-ary to explore the extreme conditions, for which were predictednsatisfactory results. After performing the 15 experiments sug-ested by this design, the two responses (R1-% relative recovery and2-% relative purity) were calculated based on the intensity of elec-rophoretic bands, as presented on Table 2. All statistical analysisere performed by using both UNICORNTM 6.1 and Design Expert

.0 software. The resultant regression Eqs. ((1) and (2)) provided theevel of the two responses (R1 and R2) as function of the differentariables, such as binding and elution conditions and temperature.

1 = +80.44 − 37.68A + 8.22B + 15.21C − 7.12AB − 3.78AC

+ 0.71BC − 30.11A2 − 7.65B2 − 5.95C2 (1)

2 = +25.54 − 0.29A + 2.73B + 3.48C − 1.37AB + 6.95AC

− 0.81BC − 7.40A − 2.44B2 − 7.56C2 (2)

To test the regression equation and check the significance of the

odel it was performed the analysis of variance (ANOVA), as it is

hown in Table 3. Through ANOVA, the quadratic regression modelemonstrated that the model is highly significant. The signifi-ance probability (P-value) was considered to be significant when

.66 0.0010 0.15 0.98 0.94 15.45

.40 0.020 0.98 0.93 0.80 9.36

lower than 0.05, when comparing modellable and unmodellablevariance. Because in both responses P-value was lower than 0.05,the data are statistically significant. According to Design Expert7.0 analysis, in % relative recovery, the model F-value of 27.66implies that the model is significant. There is only a 0.10% chancethat a “Model F-Value” this large could occur due to noise. In thesame way, in % relative purity, the model F-value of 7.40 impliesthat the model is significant. There is only a 2.01% chance that a“Model F-Value” this large could occur due to noise. The Lack of Fitexpresses if the model is adequate to describe the observed dataor if a more complicated model should be used. This parameteris achieved by comparing the variability of the current modelresiduals to the variability between observations and replicatesettings of the factors. Therefore, as the Lack of Fit of P-value wasfound to be non-significant, inferior to 0.05, for both responses,it suggests that the model equation was adequate to predict thepercentage of relative miRNA recovery and purity under any setsof the variables combination [29]. The explained variation, R2, hasto be between 0 and 1, with larger values being more desirable.Here, the relative recovery value (0.98) was better than the relativepurity value (0.93), but both represented good coefficients, whichindicated a high significance of the model [29,30]. On the otherhand, the “adjusted” R2 (Adj R2) is a variation of the ordinary R2

and is adjusted for the “size” of the model, that is the numberof the factors. The present results show that the model had agood adjusted R2 on both responses [29,30]. Moreover, the AdeqPrecision value measures the signal due to noise ratio and a valuegreater than 4 is desirable. The % relative recovery and % relativepurity have 15.45 and 9.36 Adeq Precision, respectively, indicatingan adequate signal. So, this model can be used to navigate thedesign space. Statistical analysis also revealed that a good modelwas obtained, because the model validity, which reflects if theright type of model was chosen from the beginning in the problemformulation, was above 0.25 for both response (0,53 for R1 and 0,99for R2) [30]. Therefore, according to ANOVA results the model fittedthe data and it was able to predict and optimize the responses.

3.2.1. Three-dimensional response surface plotsWith the three-dimensional response surface plot and contour

plot for the % relative recovery (R1) and % relative purity (R2)(Figs. 1 and 2, respectively) it was possible to see that the factors areclosely related. For the percentage of recovery, the contour plot ofbinding versus temperature and of elution versus binding showeda visible ellipticity. It was clear that the interaction between thetwo variables occurs. On the other hand, elution and temperaturefactors do not show interaction, which means that temperature ismore important when related to binding condition. For the percent-age of purity, it was observable ellipticity or in all graphics, althoughthe elution versus temperature displayed a weaker interaction.

3.2.2. Model validationWith the analysis of all responses on Design Expert 7.0 soft-

ware, it was possible to predict the best conditions leading tothe best predicted responses. These conditions needed to reflect

the most pure and concentrated pre-miR-29 peak, so the purityachieved was favored toward the recovery. The optimum condi-tions obtained are described in Table 4. To better understand whereoptimum conditions stand on design space, Fig. 3 presents two

A. Afonso et al. / J. Chromatogr. A 1343 (2014) 119–127 123

Fig. 1. Contour plot (A) and three-dimensional response surface plot (B) of interactions of variables binding, elution and temperature and its effect on % relative recovery(R1). One parameter for each graph is at a hold value, corresponding to the central point of the absent variable.

124 A. Afonso et al. / J. Chromatogr. A 1343 (2014) 119–127

Fig. 2. Contour plot (A) and three-dimensional response surface plot (B) of interactions of variables binding, elution and temperature and its effect on % relative purity (R2).One parameter for each graph is at a hold value, corresponding to the central point of the absent variable.

A. Afonso et al. / J. Chromatogr. A 1343 (2014) 119–127 125

Fig. 3. Contour plots for the predicted responses of the defined optimal point. (A) Predicpurity response.

Table 4Optimum conditions settled by DoE for pre-miR-29 purification.

Factor Value

Binding ([(NH4)2SO4] in 10 mM Tris buffer, pH 8) 1.70 MElution ([(NH4)2SO4] in 10 mM Tris buffer, pH 8) 1.11 MTemperature 16.11 ◦C

Predicted responses Value

cedpb

FT

% Relative recovery 84.24% Relative purity 26.61Desirability 0.90

ontour graphics with the two predicted responses, obtained with

stablished optimum conditions. Furthermore, according to theseata, the optimum gradient (1.7 M (NH4)2SO4 in 10 mM Tris buffer,H 8; 1.11 M (NH4)2SO4 in 10 mM Tris buffer, pH 8; 10 mM Trisuffer, pH 8) at 16.11 ◦C was repeated three times (Fig. 4(A)) in

ig. 4. (A) Chromatographic profiles of three replicates using the optimized stepwise grris–HCl, pH 8; 10 mM Tris–HCl, pH 8. (B) Polyacrylamide gel electrophoresis of the peak

ted percentage of relative recovery response. (B) Predicted percentage of relative

order to evaluate the reproducibility and suitability of the method.All chromatograms (Fig. 4(A)) revealed the same profile, suggest-ing that the same interactions occur in all experiments. For furtheranalysis, peak fractions were examined with polyacrylamide gelelectrophoresis (Fig. 4(B)) and two different responses were deter-mined. The % relative recovery and the % relative purity wereestablished for 1.11 M (NH4)2SO4 resulting peaks, because it wasexpected that the pre-miR-29 elution occurred with the decreaseof the ionic strength.

The responses obtained from optimization/validation of themodel were in accordance with the chromatograms. The mean ofvalues for % relative recovery was 83.3% with a standard deviationof 3.14% and for % relative purity was 24.5% with a standard devia-

tion of 1.45%. The variation between the responses was acceptablebeing between the confidence interval (CI). The 95% CIs representthe range in which responses should lie 95% of the time, and pre-vious results are within the expected [31]. With Design Expert 7.0

adient of 1.7 M (NH4)2SO4 in 10 mM Tris–HCl, pH 8; 1.11 M (NH4)2SO4 in 10 mM fractions of Run 1, 2 and 3.

126 A. Afonso et al. / J. Chromatogr. A 1343 (2014) 119–127

F (NHT injectep

d7tuo

3

bt(ittbt2dewilltiiitp

3

b

ig. 5. (A) Chromatographic profile of RNAs elution with a stepwise gradient of 1.6 Mris–HCl, pH 8. (B) Polyacrylamide gel electrophoresis of the peaks 1–3. (C) Lane S—roducts. Lanes 1–3—PCR products obtained using the fractions 1–3 as templates.

ata, the 95% confidence interval for the % relative recovery was2.6% to 95.9% and for % relative purity was 21.95% to 31.28%. Thus,hese results confirm that the model fitted well the data, it beingseful to predict the effect on responses when any factor is changed,n design space.

.2.3. Temperature effectAfter the design responses (Table 2) analysis, it was possi-

le to conclude that the temperature factor has a critical role onhe pre-miRNA retention. Comparing the results at 4 ◦C and 20 ◦Cextremes), the eluted pre-miRNA at 20 ◦C was more pure and alson higher concentration. In this work it was verified that at a loweremperature, the RNAs were strongly attached to the column andhat the temperature of 16.11 ◦C promoted the right interactions toind the RNAs to matrix and to selectively elute pre-miR-29. Whenhe conductivity decreased, pre-miR did not elute as occurred in0 ◦C experiments, as it is clear on Table 2 in Runs 9–10 and 11–12ata. Although no studies have been done toward temperatureffect in RNA and O-phospho-l-tyrosine interactions, the presentork revealed that at a lower temperature, the binding strength

s increased showing the same behavior already described in theysine matrix [26]. Here, the RNA binding was favored even atower temperatures, confirming the presence of additional interac-ions, associated with the exposure of biomolecules to the ligands,ncreasing their retention on the matrix. Because it was clearly ver-fied that temperature influenced the retention of different RNAs,t was important to choose this parameter in the process optimiza-ion. Working at the right temperature (16.11 ◦C) conducted to besturification results.

.3. Final optimization

The information given by BBD was used in order to get theest results from the purification process. Besides that, it was

4)2SO4 in 10 mM Tris–HCl, pH 8; 1.11 M (NH4)2SO4 in 10 mM Tris–HCl, pH 8; 10 mMd sample. Lane M—DNA molecular weight marker. Lane C—negative control of PCR

verified that the optimum point compromised the purificationdegree, as the quantity of miRNA recovered increased. Therefore, asmall rearrangement in conditions already optimized could lead tobetter results and some new experiments were performed. In fact,the elution condition of 1.11 M (NH4)2SO4 in 10 mM Tris buffer(pH 8) after a 1.6 M (NH4)2SO4 in 10 mM Tris buffer (pH 8) bindingresulted on a purest miRNA (Fig. 5). These conditions were exploredbecause the previous range of factors could be affecting the results,so similar conditions were investigated, to see if pre-miRNA puritywould improve. By analysis of Fig. 5(B), it was verified that thesenew conditions resulted in 52% of pure pre-miRNA and a relativerecovery of 71%. Because the salt concentration of the bindingcondition was decreased, small RNAs, which were contaminatingthe pre-miR-29 peak, eluted immediately after injection. In thisway, the step of 1.11 M (NH4)2SO4 was capable of eluting purepre-miR-29 as the others contaminants eluted only on the Trisbuffer step. With cDNA determination (Fig. 5(C)), it was verifiedthat the pre-miRNA did not elute on the first peak. These resultsconfirmed that most pre-miR-29 was eluted on the second peak,although a small amount remained retained on the column, itbeing only eluted after a decrease of conductivity. The small lossof pre-miR-29 on the third peak is acceptable, because it wasverified that in order to have a more pure biomolecule on thesecond gradient step, it was necessary to decrease the percentageof recovery toward the percentage of purity.

4. Conclusions

In this study, the pre-miRNA was obtained from a recombinanthost and the Box–Behnken design was used to optimize the purifi-

cation process. The design tool tested three factors (binding andelution conditions and temperature) at three levels, with two repli-cates of the central point. Although the result (83% of recovery and25% of purity) was close to the model prediction, the final obtained

atogr.

pTcmHbpcttttiermc

A

fBFaÂ(tFtDTaa

R

[[[[[[

[

[[[[

[

[

[[[[[[

[

[30] E. Johansson, L. Eriksson, N. Kettaneh-Wold, C. Wikstrom, S. Wold, Design of

A. Afonso et al. / J. Chrom

urity (52%) was improved after testing similar conditions.herefore, it was possible to verify that the purification degree wasompromised by the design space, because when the quantity ofiRNA recovered increased, the quantity of pure miRNA decreased.ence, and considering the therapeutic application of the targetiomolecule, “one at a time” method could be useful to manage theurification in detriment of some recovery. It is possible to con-lude that design of experiments could be used as a faster approacho initially obtain the optimum conditions, revealing the impor-ance of factors and the complexity of the problem. Additionally,his optimizing approach could bring information for first predic-ions concerning optimal factor settings and establishing new pathsn the final purification. As chromatographic processes are influ-nced by many variables, it is hard to optimize and guarantee theeproducibility of all responses. Therefore, the design of experi-ents (DoE) approach allows to know how the most important and

ontrollable variables affect the process and in which magnitude.

cknowledgments

This work was supported by the Portuguese Foundationor Science and Technology (FCT) by the project PTDC/EBB-IO/114320/2009 and EXPL/BBB-BIO/1056/2012 COMPETE:COMP-01-0124-FEDER-027560. Adriana Afonso acknowledges

fellowship Ref EXPL/BBB-BIO/1056/2012, Patrícia Pereira andngela Sousa also acknowledge Ph.D. and Postdoctoral fellowships

Ref SFRH/BD/81914/2011 and Ref SFRH/BPD/79106/2011, respec-ively). The authors acknowledge the program COMPETE and theCT project (Pest-C/SAU/UI0709/2011). The authors would like tohank Prof. Yo Kikuchi (Division of Life Science and Biotechnology,epartment of Ecological Engineering, Toyohashi University ofechnology) for kindly provided the pBHSR1-RM plasmid. Theuthors acknowledge to Susana Ferreira by the fruitful discussionsbout the method validation.

eferences

[1] S. Ruegger, H. Grosshans, Trends Biochem. Sci. 37 (2012) 436.

[

A 1343 (2014) 119–127 127

[2] C.Y. Chu, T.M. Rana, J. Cell. Physiol. 213 (2007) 412.[3] A.R. Meek, G.A. Simms, D.F. Weaver, J. Psychiatry Neurosci. 38 (2012)

120166.[4] S.S. Hebert, K. Horre, L. Nicolai, A.S. Papadopoulou, W. Mandemakers, A.N.

Silahtaroglu, S. Kauppinen, A. Delacourte, B. De Strooper, Proc. Nat. Acad. Sci.U.S.A. 105 (2008) 6415.

[5] N. Schonrock, M. Matamales, L.M. Ittner, J. Gotz, Exp. Neurol. 235 (2012) 447.[6] B.C. Bernardo, F.J. Charchar, R.C. Lin, J.R. McMullen, Heart Lung Circ. 21 (2012)

131.[7] S.C. Tan, B.C. Yiap, J. Biomed. Biotechnol. 2009 (2009) 574398.[8] N. Nonne, M. Ameyar-Zazoua, M. Souidi, A. Harel-Bellan, Nucleic Acids Res. 38

(2010) e20.[9] R.T. Batey, J.S. Kieft, RNA 13 (2007) 1384.10] F. Sousa, D.M. Prazeres, J.A. Queiroz, Trends Biotechnol. 26 (2008) 518.11] A. Sousa, F. Sousa, J.A. Queiroz, J. Chromatogr. B 879 (2011) 3507.12] A. Sousa, C.T. Tomaz, F. Sousa, J.A. Queiroz, J. Chromatogr. A 1218 (2011) 8333.13] R. Martins, J.A. Queiroz, F. Sousa, J. Mol. Recognit. 23 (2010) 519.14] R. Martins, J.A. Queiroz, F. Sousa, Biomed. Chromatogr. 26 (2012) 781.15] P. Pereira, A. Sousa, J. Queiroz, I. Correia, A. Figueiras, F. Sousa, J. Chromatogr. B

951–952 (2014) 16.16] P. Pereira, A. Sousa, J. Queiroz, A. Figueiras, F. Sousa, J. Chromatogr. A 1331

(2014) 129.17] Y. Fukami, F. Lipmann, Proc. Nat. Acad. Sci. U.S.A. 82 (1985) 321.18] W. Filipowicz, Cell 122 (2005) 17.19] F. Frank, N. Sonenberg, B. Nagar, Nature 465 (2010) 818.20] J.B. Ma, Y.R. Yuan, G. Meister, Y. Pei, T. Tuschl, D.J. Patel, Nature 434 (2005)

666.21] M.H. Rafamantanana, B. Debrus, G.E. Raoelison, E. Rozet, P. Lebrun, S. Uverg-

Ratsimamanga, P. Hubert, J. Quetin-Leclercq, J. Pharm. Biomed. Anal. 62 (2012)23.

22] S.L. Ferreira, R.E. Bruns, E.G. da Silva, W.N. Dos Santos, C.M. Quintella, J.M. David,J.B. de Andrade, M.C. Breitkreitz, I.C. Jardim, B.B. Neto, J. Chromatogr. A 1158(2007) 2.

23] G. Guiochon, L.A. Beaver, J. Chromatogr. A 1218 (2011) 8836.24] H. Suzuki, S. Umekage, T. Tanaka, Y. Kikuchi, J. Biosci. Bioeng. 112 (2011) 458.25] N. Schonrock, J. Gotz, Cell. Mol. Life Sci. 69 (2012) 3543.26] A. Sousa, F. Sousa, J.A. Queiroz, J. Sep. Sci. 33 (2010) 2610.27] F. Sousa, C. Cruz, J.A. Queiroz, J. Mol. Recognit. 23 (2010) 505.28] S.L. Ferreira, R.E. Bruns, H.S. Ferreira, G.D. Matos, J.M. David, G.C. Brandao, E.G.

da Silva, L.A. Portugal, P.S. dos Reis, A.S. Souza, W.N. dos Santos, Anal. Chim.Acta 597 (2007) 179.

29] D.C. Montegomery, Design and Analysis of Experiments, fifth ed., John Wiley &Sons, Inc., United States of America, 2001.

Experiments: Principles and Applications, third ed., MKS Umetrics AB, Sweden,2008.

31] R.F.G. Robert, L. Mason, JamesL. Hess, Statistical Design and Analysis of Exper-iments, second ed., John Wiley & Sons, Inc., New Jersey, NY, 2003.