14 N- 1 H HMQC solid-state NMR as a powerful tool to study amorphous formulations – an exemplary study of paclitaxel loaded polymer micelles

Amorphous drug-polymer formulations are complex materials and often challenging to characterize, even more so if the small molecule component itself is increasingly complex. In this work, we present 14 N- 1 H HMQC magic-angle spinning (MAS) NMR experiments in the solid state as a promising tool to study amorphous formulations. Poly(2-oxazoline) based polymer micelles loaded with different amounts of the cancer drug paclitaxel serve to highlight the possibilities offered by these experiments: While the dense core of these polymeric micelles prevents an NMR spectroscopic analysis in solution and the very similar 15 N chemical shifts hamper a solid-state NMR characterization based on this nucleus, 14 N is a very versatile alternative. 14 N- 1 H HMQC experiments yield well-separated signals, which are spread over a large ppm range, provide information on the symmetry of the nitrogen environment and probe 14 N- 1 H through-space proximities. In this way, the overall complexity can be narrowed down to specific N-containing environments. The results from the experiments presented here represent a valuable puzzle piece, which helps to improve the structural understanding of drug-polymer formulations. It can be straightforwardly combined with complementary NMR spectroscopic experiments and other analytical techniques.


Introduction
Paclitaxel (PTX, Scheme 1) is an effective anti-cancer drug for a wide range of tumours, but it exhibits a very low aqueous solubility of 0.4 µg/mL. 1, 2 Therefore, a variety of different formulations for PTX was developed. 3 This includes the proteinbased nanoparticle Abraxane®, the polymer-conjugate Opaxio TM as well as Genexol-PM® and NK105. 4,5 The latter two contain PTX in polymer micelles formed by polyethylene glycolpolylactic acid (Genexol-PM®) and polyethylene glycolpoly(amino acid) (NK105) copolymers. In general, the development of formulations and their clinical trials require very long times with highly uncertain outcomes. The above mentioned formulations have undergone multiple phase II/III clinical trials over the past decades, e.g. a recently published phase III clinical trial of NK105, 6 and only Abraxane® is successfully marketed with FDA approval since 2007. Furthermore, they are all characterized by a relatively low PTX loading not exceeding 25 wt.%. In this context, Luxenhofer et al. reported promising preclinical data for polymer micelles comprising a poly(2-oxazoline) (POx) based triblock copolymer loaded with up to 50 wt.% of PTX. 7,8 Compared to the clinically approved PTX formulations, their formulations showed higher maximum tolerated doses, elevated drug exposure to tumour tissue and prolonged survival for mice bearing A2780 human ovarian tumours. Schulz et. al. 9 , Jaksch et al. 10 and Sochor et al. 11 studied their micellar morphology for different drug loadings (PTX and curcumin) using dynamic light scattering, atomic force microscopy, (cryogenic) transmission electron microscopy, and small-angle neutron scattering (SANS). They found that the pure polymer self-assembles into wormlike and spherical micelles in aqueous solution, while incorporation of PTX led to the exclusive formation of spherical particles. Interestingly, SANS data revealed small, PTX rich domains partly submerged within the micellar core. Apart from such morphological aspects and to be able to design improved drug delivery systems, it is very important to also understand the complex structural arrangements in such amorphous drugpolymer formulations on a molecular level. NMR probes the local environment of molecules and is thus an excellent tool to report on various possible interactions. This doesn't necessarily require crystalline or well-ordered structures, but also gives excellent results for disordered as well as amorphous samples. [12][13][14] In particular, solid-state NMR has been established as a powerful tool for structure elucidation in pharmaceutical contexts such as identification of polymorphs or investigation of amorphous solid dispersions. [15][16][17] The sometimes strongly reduced mobility of encapsulated drugs within drug formulations can hinder the characterization by NMR in solution due to strong broadening of the signals, which makes solid-state NMR the corresponding method of choice. Callari et al. as well as Pöppler et al. recently showcased how solid-state NMR at moderate to fast Magic Angle Spinning (MAS) helps to obtain loading dependent structural insights into micellar formulations with an assumed core-shell structure. 18,19 Both groups found that increasing the loading of two different types of polymeric micelles did not just affect the micellar core but interestingly also the surrounding shell, which served as a basis to explain their physicochemical and biological properties such as reduced cellular uptake and inferior dissolution rates. Hirsh et al. have recently also shown how fast 1 H MAS measurements enable the rapid characterization of general pharmaceutical dosage forms. 20 The enormous progress made with respect to hardware development, e.g. MAS frequencies of up to 130 kHz being available now, 21 also enables high-resolution detection of protons in the solid state and longer coherence lifetimes. [22][23][24] The latter are essential prerequisites for sophisticated homonuclear as well as heteronuclear 2D experiments, which are exciting tools to learn more about the intermolecular proximities in a large range of materials. However, for the solid-state NMR spectroscopic characterization of PTX and its formulations as introduced above, a series of complexities arises: PTX alone with its 51 protons and thus a variety of strongly overlapping chemical environments already poses a challenge. In addition, crystalline PTX contains two individual molecules in the asymmetric unit (Z'=2), 25 which leads to a doubling of the expected resonances. Consequentially, up to now, only 13 C solid-state NMR data with an assignment based on comparison with structurally related fragments and calculations has been published, 26,27 while 1 H solid-state NMR data of PTX is still missing. Including the polymer component and potential drug-polymer interactions into the picture further complicates the spectral evaluation and thus requires narrowing down the search space, e.g. by focusing on specific functionalities. For example, spectral simplification can be achieved by making use of NMR active heteronuclei. The frequent involvement of nitrogen atoms in hydrogen bonding in pharmaceuticals makes this nucleus a promising starting point for a detailed investigation of PTX containing formulations through 2D nitrogen-proton correlation experiments, especially since PTX has only one nitrogen environment per molecule. There are two NMR-active nitrogen isotopes with low magnetogyric ratios: 15 N (I = 1/2) possesses a low natural abundance of only 0.37 %, whereas 14 N has a natural abundance of 99.6 % but exhibits quadrupole broadening due to its nuclear spin quantum number I = 1. The substantial disadvantage of the first-order quadrupole interaction, causing a broadening in the range of several MHz, can be avoided by setting the indirect 14 N spectral width of the 2D correlation equal to the spinning frequency. Tatton et. al. successfully applied this in the form of a 14 N-1 H heteronuclear multiple-quantum correlation (HMQC) experiment to study crystalline model compounds, 28 hydrogen bonding in cocrystals as well as attempting a first transfer to amorphous solid dispersions. 29 So far, the majority of 14 N-1 H HMQC experiments was acquired for highly ordered, crystalline compounds. 28,[30][31][32][33][34][35][36][37][38][39] In contrast, disordered systems with significant differential dynamics are more challenging systems and there are thus very seldom reports employing the 14 N-1 H HMQC experiment for their characterization. However, the characterization of disordered and amorphous materials could substantially benefit from the additional 14 N second-order isotropic quadrupolar shifts (see SI for detailed equations) as opposed to the 15 N isotropic chemical shifts alone. The additional 14 N second-order isotropic quadrupolar shifts will result in the spreading of the nitrogen signals over a larger ppm range (hundreds of ppm) and can thus disperse very similar or even overlapping 15 N NMR signals. This could prove to be crucial for the analysis of amorphous drug formulations, either when the nitrogen in both, drug and polymer, is part of the same functional moiety (e.g. amides in the previously mentioned POx/PTX formulations) or when the distribution of similar local environments as found in amorphous forms makes the differentiation between peaks challenging. This is underlined by the fact that around 84% of unique small-molecule drugs contain a nitrogen atom. 40 As a result, there is a large, yet largely unexplored potential of this experiment for the analysis and increased understanding of such materials. In solid-state NMR, there are two different ways to create 14 N-1 H HMQCs, either by heteronuclear through-bond 14 N-1 H J-couplings and residual second-order quadrupole-dipolar couplings or by 14 N-1 H through-space dipolar couplings. 33,34,41 By varying the recoupling time in the latter case, it is then possible to obtain different through-space proximities with longer recoupling times probing longer range intra-and intermolecular N···H distances. 28 A practical guideline with a focus on experimental procedures and parameters can be found in recent literature. 42 Combining all aspects, 14 N-1 H HMQC experiments represent a very promising tool for the investigation of the complex drugpolymer assemblies. Scheme 1: Structural formula of the components used in this study: The amphiphilic block copolymer poly(2-methyl-2-oxazoline)-block-poly(2-n-butyl-2-oxazoline)-blockpoly(2-methyl-2-oxazoline) (POL) encapsulates paclitaxel (PTX) by self-assembly into polymeric micelles (schematic drawing on the right).
In the present work, we investigate the suitability and potential of 14 N-1 H solid-state NMR experiments for the analysis of amorphous polymer-PTX formulations, for which structural information is challenging to obtain due to the lack of longrange order. Therefore, differently loaded paclitaxel formulations based on the amphiphilic triblock copolymer poly(2-methyl-2-oxazoline)-block-poly(2-n-butyl-2-oxazoline)block-poly(2-methyl-2-oxazoline) (pMeOx-b-pBuOx-b-pMeOx = Please do not adjust margins Please do not adjust margins A-pBuOx-A = POL) (Scheme 1) serve as the set of samples. 8 Complemented by samples of the individual components, the following points will be addressed: (i) general feasibility of the experimental setup for the individual components, (ii) exploring the potential of 14 N-1 H HMQC experiments for the analysis of their amorphous micellar formulations, (iii) extracting information on 14 N-1 H proximities for intermolecular throughspace contacts and (iv) their interpretation with respect to (loading dependent) structural features as a complementary source of information to existing data obtained by techniques such as SANS.

Materials
Paclitaxel in its crystalline form was purchased from LC Laboratories and used without further purification. The ABA triblock polymer poly(2-methyl-2-oxazoline)-block-poly(2-nbutyl-2-oxazoline)-block-poly(2-methyl-2-oxazoline) (POL) was synthesized according to previously published protocols. 43 Subsequent preparation of the differently loaded PTX-POL formulations also followed the literature known procedures. 8,9 A short description can be found in the supporting information. Resulting drug loadings were determined using HPLC analysis. The corresponding micellar formulation denoted as POL-2-PTX, POL-4-PTX and POL-9-PTX contain 17 wt % (10:2), 29 wt % (10:4) and 47 wt % (10:9) PTX. 14 N-1 H HMQC experiments were performed using a Bruker Avance III spectrometer at a 1 H Larmor frequency of 850 MHz. A Bruker 1.3 mm triple resonance probe operating in double resonance mode at a MAS frequency of 60 kHz was used. A pulse sequence diagram is shown in Figure S1. Heteronuclear dipolar couplings were reintroduced by applying rotary resonance recoupling (R³) 44 at the = 2 condition, proposed by Gan et. al. 33 , using x,-x phase inversion 45 . Each recoupling block had a length of 16.67 µs. The durations of the 1 H pulses/ 14 N pulses were 1.6 µs/3.2 µs for PTX and POL-9-PTX, 1.55 µs/3.1 µs for POL-2-PTX, POL-4-PTX, and 1.55 µs/3.2 µs for POL. A fourstep nested phase cycle was used to select changes in coherence order Δ = ±1 (on the first 1 H pulse, two steps) and Δ = ±1 (on the final 14 N pulse, two steps). Correct calibration of the magic angle is crucial for this experiment. All NMR data including the compiled pulse sequences for the results presented in this paper can be found in the Warwick Research Archive Portal (WRAP). Magic angle and pulse calibrations were done with the dipeptide β-AspAla. 14 N chemical shifts were referenced to a saturated NH4Cl at -352.9 ppm, corresponding to a primary reference of CH3NO2 at 0 ppm. 15 N experiments recorded at 9.4 T were referenced to unlabelled α-glycine at -342.0 ppm and those at 20 T were referenced to labelled Lhistidine, which has an NH3 + resonance at -333.1 ppm. Both ways of 15 N referencing correspond to a CH3NO2 reference at 0 ppm. The 13 C NMR data were measured with a Bruker Avance III HD 600 MHz spectrometer and a 3.2 mm double-channel probe.

NMR
The duration of the 1 H pulses was 2.5 µs in the 13 C CP MAS with ramped cross-polarization conditions optimized using a ramp from 90 to 100 with 100 increments and the sample α-glycine. SPINAL-64 46 heteronuclear decoupling was applied during an acquisition period of 23 ms at a 1 H rf nutation frequency of 100 kHz, with an optimized pulse length of 4.9-5.1 µs. The 15 N data of POL-9-PTX were recorded with a Bruker Avance III spectrometer at a 1 H Larmor frequency of 850 MHz and a duration of the 1 H pulses of 2.5 µs with optimized crosspolarization conditions using histidine. The 15 N data of crystalline PTX and POL were performed using a Bruker Avance Neo 400 MHz spectrometer. In both cases, 4 mm rotors with a sample volume of 80 μL were used). The duration of the 1 H pulses was 2.5 µs with optimized cross-polarization conditions using α-glycine. KBr was used for magic angle calibration. 13 C Chemical shifts were referenced to the methylene carbon of adamantane at 38.48 ppm. 47

Results
As a starting point for the spectral assignment and further investigation of the three differently loaded formulations of the poorly water-soluble cancer-drug paclitaxel containing 17 wt.% (POL-2-PTX), 29 wt.% (POL-4-PTX) and 47 wt.% (POL-9-PTX) PTX, NMR experiments in solution were performed (Chapters SI 4-6). For the assignment of PTX, data from NMR measurements in CDCl3 were used. However, in aqueous polymer micellar formulations, only extremely broad, unresolved PTX signals can be observed in the 1 H NMR spectra (Chapter SI 6 with Figure S4), underlining the necessity to focus on solid-state NMR for the analysis of such formulations. Consequently, a set of different solid-state NMR spectra at moderate to fast MAS spinning frequencies was recorded for the formulations and as-received PTX, amorphous PTX and the neat polymer. Both the 13 C CP MAS NMR spectra recorded at 24 kHz ( Figure  1a) as well as powder X-ray diffraction data ( Figure S5) confirm that the formulations are amorphous materials in agreement with a single glass transition temperature published in a previous work by one of the authors. 9 While crystalline paclitaxel (black spectrum) shows two sets of carbon resonances due to its two independent molecules in the asymmetric unit, amorphous PTX and an exemplary formulation (POL-9-PTX, highest PTX content) (grey and blue spectrum) show broad signals as expected for materials lacking long-range order and containing a distribution of environments and thus chemical shifts. Consequently, several moieties of PTX appear as joint signals complicating a straightforward extraction of spectral changes upon incorporation into the micelles in a similar way as recently shown for related polymer micelles loaded with curcumin as model compound. 19 Moreover, additional difficulties arise from overlap between PTX and POL resonances as can be seen from comparison with the 13 C CP MAS spectrum of the neat polymer (red spectrum). Note, that for better comparability, the neat polymer was heated above its glass transition temperature Tg (Tg(POL)=56°C) 48 prior to the measurement to account for similar conditions during the preparation of the formulations. Consequently, the analysis of the carbon spectra and extraction of reliable structural information is difficult and requires complementary tools such as quantum chemical calculations, which will be explored in detail in a separate work.
Ideally, for a thorough characterization of the formulations we would like to use the direct information from 1 H NMR data at fast MAS due to the high sensitivity of the 1 H nucleus to intermolecular interactions, through space proximities and subtle changes in the packing arrangement. The 1 H MAS NMR spectra recorded at 60 kHz are depicted as external projection in the corresponding 14 N-1 H HMQC spectra in Figure 2 and Figure 3. As can be seen, the spectra are dominated by severe signal overlap despite being recorded at high magnetic field (20 T). Consequently, spectral simplification is required and the presence of only one secondary amide moiety per PTX enables this through N-H correlation experiments. Additionally, the tertiary amide functional groups in the polymer could also serve as another source of structural information. Therefore, 15 N CP MAS spectra were recorded and it is very interesting to note that the secondary amides from PTX (Z' = 2, crystal structure in SI 10) and the tertiary amide of the polymer show resonance in the same spectral region (Figure 1b). For the formulation POL-9-PTX, only a very broad resonance can be observed, again in a similar spectral region and thus showing the need for a dispersion of the signals as expected for the quadrupolar 14 N nucleus. 14 N-1 H HMQC MAS NMR experiments are particularly powerful in this context, require only a small amount of sample and have been applied to study a variety of crystalline compounds. 33,35 Therefore, in a first step, 14 N-1 H HMQC spectra of the pure compound were recorded to identify their nitrogen shifts and ensure sufficient separation of the PTX and polymer nitrogen environments.  Figure S6 and Tatton et. al. 28 ). For PTX, using short recoupling times and in agreement with the X-ray diffraction data, two cross peaks at 8.8/-30 ppm (N A ) and 8.0/-18 ppm (N B ) were observed for the NH groups of the two individual molecules of PTX in the asymmetric unit. Consequently, one can deduce that N A is involved in stronger hydrogen bonding than N B . This agrees with a longer direct N A •••H distance, e.g. stronger involvement in hydrogen bonding, as observed in the crystal structure reported by Vella-Zarb et. al for PTX anhydrate (Table  1, CSD code: RIGLEA). 25 With a three times higher recoupling time, two changes of the previous peaks for shorter recoupling time are noticeable: The intensity of the peak corresponding to nitrogen N B increases compared to N A , the peak seems elongated in the 14 N dimension and two additional signals around 5 ppm proton chemical shift are observed. Small changes in the 14 N shift (N A : -30 to -40 ppm and N B : -18 to -22 ppm) could be caused by a temperature increase in the sample caused by the longer irradiation through the recoupling blocks. As the recoupling time for the dipolar couplings increases, it is possible to sample longer N•••H distances through space. Consequently, the relative signal increase of N B can be explained by proximities of N A and N B to aromatic protons, whose chemical shifts are overlapping with the proton attached to N B in agreement with an increased 1 H intensity at 8.0 ppm. To confirm this, intra-and intermolecular close 14 N••• 1 H distances up to 3 Å were extracted from PTX  comparable recoupling times. 28 Indeed, the shortest contact for N1 is found to the ortho-CH of the adjacent aromatic ring denoted as carbon 32 in the SI. In turn, the shortest NH distances for the second molecule were observed to adjacent aliphatic CH and OH units (C3´H and C2´OH), which could explain the cross peaks observed at 5.0 and 5.3 ppm in the 1 H dimension. However, the distances in Table 1 can only serve as indication and should be viewed carefully as these distances originate from PXRD data at high temperature (360 K). Furthermore, it is still not fully understood why, for some small molecule compounds, fewer contacts than expected are observed. 42 In a 14  Peak B indicates proximity to the CH2 unit of the hydrophobic butyl sidechain (C17/18) and peak C is related to N···H proximities including the methyl protons of the butyl sidechain (C19). Please note, that due to the micelle formation, these cross-peaks are most likely of intermolecular nature. Interestingly, no cross peaks between N and the polymer backbone CH2 groups could be observed. For a more detailed discussion on the signal assignment and appearance of specific cross-peaks, the reader is referred to the SI 4. Overall, the comparison of the two individual components shows that their nitrogen environments can be clearly distinguished in the 14 N-1 H HMQC experiment, which is not the case if 15 N chemical shifts are observed (see Figure 2b and Table 1). Additionally, distinct correlation peaks are obtained in the 2D NMR spectra significantly reducing the spectral complexity in the 1 H dimension. This is an essential prerequisite for the following analysis of the formulations. While revealing valuable intra-and intermolecular N•••H proximities, the HMQC experiments also feature considerably shorter experimental times (1h 30 min up to 7 h) than the 1D 15   To complete the set of differently loaded formulations and subsequently extract trends, which might reveal information on changes of the local environment in the formulations, 14 N-1 H HMQC data was also recorded for POL-9-PTX, a formulation with almost 50 wt.% PTX loading (Figure 4). Due to hardware  All extracted values for both the individual components and the formulations are summarized in Table 1, which also includes selected 15 N chemical shifts extracted from the spectra in Figure  1b. This first examination of the 14 N-1 H HMQC spectra of the three formulations is very promising as two different nitrogen environments can be distinguished for each sample, which follow a clear trend of decreasing 14 N shift upon increasing loading. An analogous trend could not have been observed based on 15 N solid-state NMR data as indicated by the similarity of the extracted chemical shifts for the polymer and the formulation with highest PTX loading.

Discussion
In a next step, the observed trends and key N•••H proximities have to be analyzed and transferred into chemical knowledge to improve our understanding of the studied formulations on a molecular level. To do so, 14 25 the NH group and all three OH-groups act as hydrogen bond donor to adjacent PTX and water molecules suggesting that this would also be the case for PTX -polymer interactions. The polymer amide nitrogen atoms serve as hydrogen bond acceptor. From previous work on similar polymers with PTX and curcumin, 11,19,49 we know that, with increasing loading, the poorly watersoluble guest not only interacts with the more hydrophobic pBuOx polymer block, but also with the hydrophilic pMeOx units. The critical drug loading, where the corona become significantly involved appears to be between 20-30 wt.% This would explain, why we don't observe this for the lowest loading. Furthermore, steric hindrance between the different PTX interaction sites and pBuOx vs. pMeOx would result in a differing deviation from ideal hydrogen bond geometries with respect to bond distances and bond angles, while, overall, the hydrogen bonded amide nitrogen atom would be found in a more symmetric, closer to tetrahedral, environment as compared to the planar, sp 2 hybridized initial environment ( Figure 5).
This underlines that 14 N-1 H HMQC NMR experiments are not only spreading out the nitrogen NMR signals but are also a promising tool to study local symmetry and detect structure determining intermolecular interactions in complex and amorphous drug-polymer formulations. To further improve the structural understanding, these finding now need to be complemented by additional experimental and detailed theoretical data.

Summary and Conclusion
In this work, we investigated amorphous polymer formulations of the anti-cancer drug paclitaxel. For both the polymer and the drug, 15 N chemical shifts in a very similar spectral area were observed, which results in significant signal overlap, especially if amorphous samples are involved. Consequently, for this set of samples, it is very challenging to distinguish small changes in the spectra and extract trends and information on interactions of specific moieties, e.g. upon increasing drug loading of the formulations. Therefore, 14 N-1 H HMQC NMR experiments were investigated as a valuable tool to disperse signals due to the additional quadrupolar shift when observing 14 N instead of 15 N. We could show that such experiments can be a rich source of information for the characterization of molecular interactions in amorphous polymer-drug nanoformulations, which is otherwise difficult to obtain. On the one hand, the 14 N signals were dispersed so that two distinct 14 N environments could be extracted for each formulation and the complexity in the 1 H dimension could be reduced through this correlation experiment. Thus, it was possible to extract trends for different drug loadings, distinguish cross peaks and extract quadrupolar parameters, which together enabled us to learn about the local amide environment, its symmetry and potential interaction motifs between drug and polymer in such complex, amorphous mixtures. Overall, the 14 N-1 H HMQC NMR experiments have great potential for the analysis of disordered and amorphous drug-delivery systems, which can be transferred and should be explored for other systems. Here, the 14 N-1 H experiments could also be insightful for nanocrystalline materials such as found for Pluronic F127 stabilized PTX nanocrystals. 50 Of importance to stress for the broader applicability of the 14 N-1 H HMQC experiment: While fast MAS (≥ 60 kHz) is essential for these experiments due to 1 H signal narrowing and most importantly increased 1 H coherence lifetimes, 28 the experiments are not limited to high magnetic fields. With the second order quadrupolar interaction being inversely proportional to the strength of the magnetic, narrower lines are obtained at higher field. However, a larger dispersion of the signals is counterbalancing this at lower magnetic field strengths. In fact, many examples for the application of this and other 14 N based experiments in literature show their feasibility at 400-600 MHz. 30,34,36,51 Furthermore, in a next step we intend to combine these experiments with additional data and complementary calculations to learn more about these PTX formulations and ultimately derive realistic structural models on a molecular level. In this context, molecular dynamics simulations could provide important insights due to the possibility to reproduce distributions of environments as encountered for amorphous drug-polymer formulations.

Conflicts of interest
There are no conflicts to declare.