When it is too much: Identifying butamben excess on the surface of pharmaceutical preformulation samples by Raman mapping

In nanostructured lipid carriers (NLC), the type and amount of excipients will determine API solubility and therefore the maximum drug load. Butamben is a topical local anesthetic which formulation in lipid-based DDS is difficult due to its affinity for hydrophilic solvents, which might not miscible with the solid lipid. This indicates that a medium polarity excipient might be needed. The first step of this study comprised a throughout screening study to evaluate API solubilization in different excipients. Then excipients with low (Dhaykol ® 6040 LW) and high (Capryol ® 90) solubilization capacities were selected for microscopic evaluation by Raman mapping. For Capryol ® 90 a mixture design of experiments was carried out to study the proportions of excipients, using as responses the DHI (distributional homogeneity index) and standard deviation of the histograms. Clusters of the API were observed I the samples prepared with Dhaykol ® , confirming the low solubilization capacity. In this case, DHI was an adequate parameter to indicate solubilization/miscibility. In the case of Capryol ® 90 samples, no clusters were observed due to its higher solubilization capacity, however since it was homogeneously distributed throughout the analyzed surface, the DHI values were low, indicating the need for a 3D image.


INTRODUCTION
Butamben (butyl 4-aminobenzoate, BTB) is a poor soluble (140 mg/L water) ester type local anesthetic (Cereda et al., 2017) used in topical preparations. Several works report its local anesthetic action, sometimes associated with tumor treatment to alleviate pain (Kim et al., 2011;Rampaart et al., 2008). Drug delivery systems (DDS) comprise several kinds of formulations aimed to enhance the therapeutic effect and/or to reduce the toxicity of active pharmaceutical ingredients (API) (Tiwari et al., 2012). Lipid-based DDS are very interesting carriers to deliver hydrophobic API. Studies of butamben encapsulated in lipid DDS such as liposomes and nanostructured lipid carriers (NLC) (Cereda et al., 2013;Maestrelli et al., 2010;Mura et al., 2021;Rodrigues da Silva et al., 2021) showed that BTB could be an interesting model for miscibility studies.
NLC are promising DDS to API of low aqueous solubility (i.e. BCS classes II and IV). The NLC core is composed by a mixture of solid and liquid lipids that retains solid features at room temperature. The most used NLC-lipid excipients are triglycerides, mixtures of partially digested glycerides, fatty acids, and typical pharmaceutical waxes.
These excipients also help to decrease the toxicity in this kind of formulation since they present the GRAS ("generally recognized as safe") status. (Williams et al., 2013). BTB, which partition coefficient (logP) is 2.87 (United States Environmental Protection Agency -EPA, 2020) is a challenging API to be uploaded by NLC. According to Brewster et al. (2007) logP values define two categories of poorly water-soluble drugs: (i) grease-ball and (ii) brick dust compounds (Brewster et al., 2007). The first have a higher logP ( > 5) and are very lipophilic compounds, with lower melting point. The last, "brick dust" compounds (logP < 3) show low solubility in lipid excipients. Despite of that, lipid-based DDS can be successfully used to enhance BTB bioavailability, however studies about the excipients as well as their interactions with the API are necessary (Holm, 2019;Koehl et al., 2019).
Preformulation is an important stage of the rational development of pharmaceutical dosage forms (Kaur et al., 2016) where several features should be evaluated such as compatibility between drug and excipients, stability, storage conditions and miscibility, especially for liquid and semi-solid formulations (Krupa et al., 2015;Lau, 2001;Renuka et al., 2017). Miscibility is related to the identity of excipients and to their concentrations. Excess amounts of drug in lipid-core DDS such as NLC and solid lipid nanoparticles (SLN) can result in the expulsion of the API from the matrix and, thus, formulation instability (Müller et al., 2016). Beyond the compounds themselves, their proportions and interactions are crucial to understand the miscibility.
Raman mapping can be a useful technique in studies of semi-solid lipid formulations. Sacré et. al. characterized a semi-solid self emulsifyig drug delivery system (SEDDS) formulation using Raman mapping in relation to its particle size and distribution (Sacré et al., 2015). Our research group has shown that Raman mapping is a powerful tool to evaluate the miscibility between lipid excipients (Breitkreitz et al., 2013;Mitsutake et al., 2018Mitsutake et al., , 2019 and between excipients and API (Mitsutake et al., 2020). In the last paper, the joint information of Raman and Near Infrared were used in a multiblock approach to fully characterize structural changes of the drug and excipients during stability studies.
In this context, the aim of this study was to employ a rational screening and methodology to find an adequate combination and proportions between butamben and lipid excipients in the early stages of pharmaceutical development. The first step was to selectby visual inspectionthe most promising lipid excipients for BTB solubilization.
Then, by using a mixture DoE and parameters obtained from Raman mapping, to determine the adequate proportion of excipients considering miscibility and drug load.

Step 1 -Screening of most promising excipients
Around 0.10 to 0.20 g of BTB was weighed and transferred to a becker, then the liquid excipient was added until BTB complete solubilization (indicated by the transparence of the solution). Solid excipients were first melted (10ºC above their melting points), and added to the BTB until complete solubilization, under stirring. In this case, the solubility of BTB was shifted to higher values, because it was measured at higher temperatures than with the lipid liquids (Jannin et al., 2015).

Step 2 -Sample preparation and Mixture Design
After the initial screening, one liquid lipid with low solubilization capacity and one with good solubilization capacity with BTB were selected. A solid excipient, cetyl palmitate (CP), was chosen to simulate the preparation of an NLC preformulation. The two chosen liquid lipids were heated 10 ºC above the melting point of cetyl palmitate, separately. Butamben was then added to the liquid lipid, under stirring, until a visually homogeneous mixture was obtained, or by five minutes in case of low solubility.
Afterwards, this mixture was poured into the solid lipid and mixed again. The samples were placed in Petri dishes and cooled at room temperature (25 ± 1 ºC). In case of low solubility with the liquid lipid, three different BTB concentrations were studied: 5.00, 14.50 and 25.00% (w/w). For liquid lipid of high solubility, a mixture design of experiments (DoE) was used.
The composition of samples in this mixture DoE is shown in Table 2. The design was a simplex lattice, with two replicates and one block. The range of concentration was 50.00 -90.00 (% w/w), 10.00 -50.00 (% w/w) and 0.00 -40.00 (% w/w) for cetyl palmitate, Capryol® 90 and butamben, respectively. Raman images acquisition was made in a random order as indicated in Table 2 to avoid bias in the results. The distributional homogeneity index (DHI) and standard deviation of histograms (STD) from Raman images were used as outputs (responses) for the mixture DoE. DHI is a parameter developed in 2014 that is employed to evaluate the homogeneity in images by a macropixels and continuous-level moving block approach (Sacré et al., 2014). Mixture DoE was created, and results analyzed using Design Expert version 11 (Stat-Ease Inc., Minneapolis, MN, USA).

Raman mapping
The samples were cooled to room temperature in an aluminum cell and an area of 4.0 × 4.0 mm (16.0 mm 2 ) was mapped using a Raman Station 400 (Perkin Elmer, CT, USA). A 785 nm laser was used as an excitation light, at 100 mW nominal power.
The exposure time was 3s/pixel and each spectrum was the average of 2 exposures. The step size was 100 µm, the spectral range was 600-3200 cm −1 with a resolution of 4 cm −1 . Each sample generated a 40 × 40 × 651 cube of data, where 40 was the number of pixels at x and y axis and 651 the number of spectral variables.

Data processing
Spikes were excluded from Raman spectra using the algorithm developed by Sabin et al. (Sabin et al., 2012). The data cube was unfolded to a 2D matrix NM × λ, where M is the number of pixels at x axis, N the number of pixels at y axis and λ the number of spectral variables. The asymmetric least squares (AsLS) algorithm was used to correct the baseline and the spectra were normalized using unit vector. The spectral range chosen to build the models was 1804-600 cm −1 .

Chemical Maps generated by Classical Least Squares (CLS)
CLS algorithm was based on the bilinear model, as shown in Equation 1: with compound concentrations and S T ( × ) contains the spectra of pure compounds.
CLS is useful in cases where there are no interactions between the compounds or spectral changes due to physical phenomena (e.g., water absorption) (Ravn et al., 2008). The chemical maps were generated by refolding C columns (Qin et al., 2019). DHI was calculated for each excipient in each map and the randomization step was repeated 100 times (Sacré et al., 2014).
The model was built using Matlab version 8.3 (Mathworks Inc., Natick MA, USA) and PLS_toolbox version 8.6.2 (Eigenvector Research Inc., Wenatchee, WA, USA). Table 3 shows the results of solubility tests at increasing solubility order. The last column shows the amount of BTB in relation to the total mass (BTB + excipients). There were excipients that could only solubilize a small amount of Butamben (as Dhaykol 6040 LW) and some in which BTB was virtually insoluble, as in the vegetable oils. The appearance of the mixtures when the API was not solubilized is shown in Figure S1A with precipitates (liquid excipients) or opaque solutions (solid excipients), while Figure   S1B shows the appearance of fully solubilized BTB formulations (clear solutions  and PEG showed better solubilization capabilities for BTB, these excipients were not employed as solid and liquid lipids, in the NLC core. Since the solid lipid was already defined as cetyl palmitate, a common excipient used in NLC preparation (Anantachaisilp et al., 2010;Rosiaux et al., 2014), the liquid lipid should be lipophilic enough to have affinity with it, i.e., to avoid phase separation. Among the excipients with high solubilization capacity, Capryol ® 90 was found to be an interesting option. Capryol ® 90, or propylene glycol monocaprylate, is a nonionic water-insoluble surfactant that can be used as co-surfactant in oral lipid-based formulations, or co-surfactant and solubilizer in topical dosage forms. Therefore, Capryol ® 90 was selected considering BTB solubility and its possible affinity with cetyl palmitate, considering its lipophilic character. Dhaykol 6040 LW was selected for comparison, as an excipient of low solubilization capability for BTB. Transcutol ® HP was not chosen because its low miscibility with the selected solid excipient, as described in previous papers (Castro et al., 2021;Mitsutake et al., 2018).

Chemical images of cetyl palmitate and Dhaykol 6040 LW samples
Raman spectra of cetyl palmitate, Dhaykol 6040 LW and butamben are shown in Figure S2 and the peak assignments are provided in Table S1. The 1700 to 1600 cm −1 spectral range could be used to build univariate maps of butamben; however, in order to better understand the relationship of lipid excipients, multivariate models were preferred. Figure 1 shows the chemical maps obtained for these samples at the initial time. As can be seen, there are no clusters of BTB at 5.0% (w/w). In the NLC structure the API tends to be solubilized in the liquid excipient, in this case, Dhaykol. This is seen in Figure 1a, by the higher BTB concentration (in red) at places where there is higher concentration of Dhaykol, i.e., the two maps are correlated. On the other hand, the cetyl palmitate map is complementary to that of Dhaykol, indicating that these two compounds are not completely miscible. At 5% w/w BTB no agglomerates were observed, and it can be concluded that the two excipients have intermediary miscibility, with standard deviation in histograms around ± 10.3 units.
The low solubility of BTB in Dhaykol resulted in API clusters at higher concentrations (Figure 1b and 1c). The chemical map of cetyl palmitate in 14.3% (w/w) BTB shows spots with absence of both the drug and Dhaykol (deep blue in Figure 1b, left and middle). In the BTB map it is clear that the API agglomerates in this region. In NLC, Dhaykol should work as a 'bridge' and it needs to be miscible with the solid lipid and to solubilize the drug, simultaneously. However, it can be clearly seen that, in this concentration, its solubilization capacity is poor and therefore there are clusters of pure BTB in cetyl palmitate. These clusters generated two populations in histograms of CP and BTB, and longer tails in Dhaykol histograms (Figure 1b). The presence of agglomerates caused a DHI increase of 1.5 or 2.0 units (Figure 1). Figure 1C shows the results for 25 % (w/w) BTB where the above-mentioned situation was confirmed. As a matter of fact, even at the visible image it was possible to see lumps ( Figure S3), that were confirmed by Raman imaging to be clusters of the drug. It is interesting to note that Raman spectroscopy allowed the detection of those clusters at the lower concentration of 14.3% (w/w) BTB, which is very interesting in preformulation studies. The miscibility problems were observed since the initial analysis and there were no alterations after 7 days (data no shown).  Figure 2 shows the chemical maps obtained from sample 11, 9, 3 and 8 (Table   2). Figure 2a     Both the DHI and STDhistograms obtained for the solid (cetyl palmitate) and liquid (Capryol) excipients were used as responses for the mixture design (Table 3). BTB results could not be used because there were samples without API, which precluded the calculation on these parameters.  Table 4 shows the model parameters for each response. Except for STDCapryol, all surfaces were described by cubic models, i.e., there were complex interactions between the compounds. All the coefficients representing the pure compounds were significant for all the responses studied. The interactions of the excipients were important for DHICapryol, the interactions of CP with BTB were important for the DHICP and STDCP models, while the interactions between butamben and CAP were important for DHICapryol. This indicates that not only the presence of API, but the interaction with the excipients affects the homogeneity/spatial distribution of the compounds in a sample.

Chemical images of cetyl palmitate and Capryol® 90 samples and Mixture DoE results
For all models the p-value for lack of fit was above 0.10, indicating that the models were well adjusted (90% confidence interval). The values of R 2 adj are a measure of the amount of variation explained by the models and they were all above 0.71. These values should be close to R 2 , but a difference of up to 0.2 is considered acceptable. However, the R 2 predicted values were very low (except for STDCapryol). These values, as well as the PRESS (predicted residual sum of squares) ones, were calculated using the leave-one-out method.
The surfaces obtained for DHICP, DHICapryol and STDCP were complex, and the removal of a single point completely changed the mathematical model, justifying the low R 2 predicted values. However, since the other parameters were adequate, the models obtained were considered representative of these samples. Table S2 shows the values of the bi coefficients of the models. The fit parameters and residuals for the DHICP are shown in Figure 4, as an example. The residuals were expected to be random, and thus, to follow a normal distribution, as shown in Figure 4a; homoscedastic, i.e., the same variance over the range     Figure 5d). This is an indication that STD is not a suitable parameter in this case. In this sense, the maximum API that could be used in this NLC is less than 5.0% (w/v).
Increasing CAP concentration could be an option to increase the quantity of BTB, as this API is soluble in this excipient. However, it is not a good alternative since the region of greatest heterogeneity occurred when the Capryol ® 90 concentration is higher than 36.67% (w/w) ( Figure 5). Therefore, although this mixture of excipients is more promising than that of CP:Dhaykol 6040 LW, the concentration of 5.0% (w/v) BTB is very close to the solubilization limit of the NLC formulation.

CONCLUSIONS
Butamben is a topical local anesthetic which formulation in lipid-based DDS is difficult due to its affinity for hydrophilic solvents, which are not miscible with cetyl palmitate, the solid lipid. This indicates that a medium polarity excipient should be used as the liquid lipid, and its miscibility with the solid lipid should be evaluated. The mixture of cetyl palmitate:Capryol ® 90 proved to be more suitable than the mixture of cetyl palmitate:Dhaykol 6040 LW, but the amount of butamben that could be added was still low. Therefore, the use of other excipientsespecially the solid lipidcould be interesting, however this was not the goal of this study.
The systematic study of excipients in the pre-formulation stage allows to find more promising excipients, saving time, cost, and materials. Raman imaging allowed not only to study of the miscibility of the excipients with the drug, but also to find regions of API overload. The ternary diagrams obtained from mixture DoE indicate the DHI value is more suitable than the STD from the histograms for miscibility studies, since in the former the information is complementary. Nevertheless, despite the vast amount of information obtained from the ternary diagrams, their evaluation should be accompanied with the evaluation of the Raman images. We also highlighted how solubility behavior can affect the sample surface and, consequently, its homogeneity. This is particularly important in the overload regions; in cases of unsuitable solubilization, heterogenous surface were generated and DHI works well as an indicative of immiscibility. However, in cases of good solubility between drug and excipient, as showed by butamben and Capryol ® 90, the good dispersion of API in the liquid lipid, caused it to be homogeneously distributed throughout the analyzed surface. Therefore, DHI values were low when they should be high. This can be an indicative of phase separation and just a surface analysis was not enough. In this sense, 3D images comparing the inner and surface part were a promising alternative to solve this problem and it is the future steps of this work.

DECLARATION OF INTEREST
The authors declare they have no conflict of interest.