Clearing the undergrowth : detection and quantification of low level impurities using 19 F NMR

A new method for the analysis of low level impurities in sparsely fluorinated species allows measurement of clean high dynamic range 19F spectra, fully decoupled and free of interfering signals from 13C isotopomers.

The high sensitivity and wide chemical shift range of 19 F NMR 1-4 make it potentially very attractive for characterising fluorinecontaining impurities. In pharmaceutical chemistry, for example, a quarter of current drugs contain one or more fluorines, 5 and regulatory authorities require all impurities above 0.1% of a main active pharmaceutical ingredient to be identified and quantified. 6 Both 1D 19 F NMR and 19 F DOSY have been used for the detection of minor fluorinated impurities. 7 One major technical problem is the difficulty of exciting quantitatively the very wide chemical shift range of 19 F, but solutions now exist for both 1D 8 and DOSY 9 experiments. However, there remains the problem of 13 C isotopomer signals. At around 0.54% of the intensity of 12 C isotopomer signals, these are in the same range as impurity signals of interest and often have similar chemical shifts, and therefore complicate their identification and quantitation. The obvious solution is to use broadband 13 C decoupling to collapse the heteronuclear J-couplings. This can work well for 1 H spectra, albeit at the expense of some sample heating. [10][11][12][13][14][15] However, 19 F is exquisitely sensitive to chemical environment and its large secondary isotope shift means that the decoupled ( 19 F-13 C) signals have slightly different chemical shifts from the parent ( 19 F-12 C) signals, so decoupling just halves the number of 19 F-13 C signals, rather than hiding them all under the parent. Here we show how to acquire clean 19 F spectra without interference from 13 C isotopomers and with no heteronuclear ( 1 H or 13 C) splittings. The new method does not use 13 C decoupling, minimising sample heating, and should greatly facilitate the detection and quantification of low-level impurities by 19 F NMR. Fig. 1 shows 19 F spectra of a slightly degraded sample of rosuvastatin (1, Scheme 1), used for treating dyslipidaemia, spiked with small amounts of precursors 2 and 3. The protondecoupled spectrum of Fig. 1a (multiplet structure renders the proton coupled spectrum, shown in Fig. S4 of the ESI, † uninformative) is complicated by the presence of both one-bond and long-range 13 C satellites; one of the two satellite signals due to the presence of 13 C at the ortho position with respect to fluorine is almost degenerate with (8 ppb from) the signal of 2.
Acquiring a spectrum with this resolution with full broadband decoupling is uncomfortably close to the limits of many instruments, because of the long high-power irradiation required, but if the one-bond 13 C satellite signals are suppressed (see Section S1 of the ESI †), low power irradiation can be used to decouple the remaining longer-range (Ztwo-bond) couplings. This gives the spectrum of Fig. 1b, in which a singlet signal is seen for the 2.2% of ortho-13 C 1. Had full 13 C decoupling been used, the ipso-13 C signal of 1, midway between the one-bond satellites in Fig. 1a, would have been degenerate with that of impurity 1a (a diastereomer). In the spectrum of Fig. 1c, in contrast, which was obtained with the new method, no resolvable signals at all are seen from 13 C isotopomers, and there is no interference with the signals of the minor components of the sample.
The new method, using the pulse sequence of Fig. 2, is compatible with several different hardware configurations; the results shown here used a single high band radiofrequency (RF) amplifier and a ( 1 H/ 19 F), 13 C triple-resonance probe with a double-tuned high band coil. The experiment consists of three parts: a low-pass filter to suppress one-bond 13 C satellite signals; a J CF -modulated spin echo; and time-shared acquisition during which the 19 F signal is recorded under 1 H decoupling.
The low-pass J filter, [16][17][18][19][20] which converts 19 F antiphase signals into unobservable heteronuclear multiple quantum coherences when D = 1/(2 1 J CF ), suppresses the one-bond 13 C satellite signals. Since a 19 F spin echo is needed to refocus the fluorine chemical shift, there is time to use two 13 C 901 pulses in a two-stage filter; if a wide range of 1 J CF values is present, further stages can be added.
The modulated spin echo, which is analogous to a heteronuclear 2D J resolved experiment, 21-23 makes the phases of the remaining 13 C satellite signals depend on the evolution time t 1 , while the desired signals from the 12 C isotopomers are unaffected. Weighted averaging of experiments with different t 1 cancels the modulated signals, leaving a clean spectrum. In practice the most effective way to perform this averaging is by double Fourier transformation and integral projection onto F 2 of the F 1 range spanned by the lineshape of the parent signal. This suppresses all satellite signals that would be resolvable in the 1D spectrum, while preserving the quantitative character of the spectrum. The final 13 C 901 pulse deals with the problem of the phasetwist lineshape 24-26 of a 2D J spectrum by suppressing the sine-modulated dispersive part of the signal. The remaining cosine-modulated signal can then be selected by zeroing the imaginary component after the first Fourier transformation, leading to signals that are doubled in F 1 but have 2D absorption mode lineshapes. The choice of increment 1/sw1 in t 1 is determined by the range of couplings to be suppressed (sw1 4 n J CH ), and the number of increments ni by the T 2 of the parent signal (ni 4 sw1 T 2 ). Relaxation losses during t 1 lead to a small sensitivity penalty for the new method, about a factor of 2 here (apparent on comparing Fig. 1a and c).
The data acquisition section of the pulse sequence uses time-shared decoupling because the 1 H and 19 F channels share the same coil in the probe used. In normal circumstances, a simple WALTZ 27,28 or similar decoupling waveform would suffice to decouple 1 H from 19 F, but the very high dynamic range of the sample means that the weak systematic signal modulations such methods induce would here give rise to significant decoupling sidebands (see Fig. S3, ESI †). These are suppressed very effectively here by the use of bilevel adiabatic decoupling. 29 As well as decoupling 1 H from 19 F during acquisition, it can be helpful to decouple in the earlier parts of the sequence, to suppress any echo modulation caused by strong 1 H-1 H coupling. This is common in aromatic spin systems (as for example in Fig. S2 of the ESI †). 21,30,31 Here the quality of decoupling is less critical, so bilevel decoupling is not needed.    Fig. 1c at which the F 2 projection of the 2D is calculated. Each 13 C isotopomer gives four symmetrically-disposed signals, with frequency coordinates (AEJ CF /2, d AE J CF /2); in Fig. S5 (ESI †) both of the less shielded satellites overlap in F 2 with t 1 -noise from the parent peak. Integration between the dotted lines produces the spectrum of Fig. 1c.
To test the quantification performance of the new method, the relative percentages of the impurities compared to the main drug substance were measured using the spectrum (Fig. S6 of the ESI †) of a fresh, undegraded, sample. Since the dynamic range of the spectrum is very high, lineshape fitting 32-36 was used instead of conventional integration. As shown in Table 1, the relative percentages measured agree well with those expected.
In systems with mutually coupled fluorines, homonuclear J modulation interferes with 13 C satellite suppression if hard 1801 19 F pulses are used in Fig. 2. Selective 1801 pulses avoid this problem, as shown in Fig. 3 for the antifungal drug fluconazole, which has J FF = 8.1 Hz. Fig. 3b and c were acquired separately using the selective analogue of Fig. 2 to excite the regions around À107 and À111 ppm respectively, revealing the degradation products 4a, 4b and 4c. 13 C isotopomer signals can pose significant challenges in identifying and quantifying impurities down to the 0.1% level. The novel approach introduced here of filtering out, rather than decoupling, these signals offers the possibility of acquiring clean, high dynamic range 19 F spectra without interference from species containing 13 C. A slightly simpler approach can be used in proton spectra.
This work was supported by AstraZeneca and by the Engineering and Physical Sciences Research Council (grant number EP/ N033949/1). All raw experimental data, and the pulse sequence code, can be downloaded from DOI: 10.15127/1.304823.  (4); (b and c) 1 H decoupled, 13 C isotopomersuppressed 19 F spectra acquired separately for each parent signal using the pulse sequence of Fig. 2 with selective 19 F 1801 pulses.