Development of a Highly Responsive Organofluorine Temperature Sensor for ¹⁹F Magnetic Resonance Applications

Abstract

Temperature can affect many biological and chemical processes within a body. During in vivo measurements, varied temperature can impact the accurate quantification of additional abiotic factors such as oxygen. During magnetic resonance imaging (MRI) measurements, the temperature of the sample can increase with the absorption of radiofrequency energy, which needs to be well-regulated for thermal therapies and long exposure. To address this potentially confounding effect, temperature can be probed intentionally using reporter molecules to determine the temperature in vivo. This work describes a combined experimental and computational approach for the design of fluorinated molecular temperature sensors with the potential to improve the accuracy and sensitivity of ¹⁹F MRI-based temperature monitoring. These fluorinated sensors are being developed to overcome the temperature sensitivity and tissue limitations of the proton resonance frequency (10 × 10⁻³ ppm °C⁻¹), a standard parameter used for temperature mapping in MRI. Here, we develop (perfluoro-[1,1′-biphenyl]-4,4′-diyl)bis((heptadecafluorodecyl)sulfane), which has a nearly 2-fold increase in temperature responsiveness, compared to the proton resonance frequency and the ¹⁹F MRI temperature sensor perfluorotributylamine, when tested under identical NMR conditions. While ¹⁹F MRI is in the early stages of translation into clinical practice, development of alternative sensors with improved diagnostic abilities will help advance the development and incorporation of fluorine magnetic resonance techniques for clinical use.

Graphical Abstract

Temperature sensing in clinical applications can provide information about altered homeostasis and progression of disease.^1,2 The body consistently regulates biochemical processes where body temperatures can fluctuate daily within 1–5 °C.^3,4 Therefore, the ability to accurately monitor variations in deep-tissue temperature could help evaluate the progression of a disease or the efficacy of heat-based therapy.^5,6 However, external temperature sensors obtain an average temperature at the local surface of the orifice of interest, making the sensors unsuitable for quantitative, noninvasive monitoring of deep-tissue internal temperatures.

In medicine, magnetic resonance (MR) thermometry is one of the most advanced systems for noninvasive body temperature measurements. Techniques have been developed to monitor the internal temperature, for example, in cases involving strokes.⁷ Temperature can be quantified by measuring the changes in nuclear magnetic resonance (NMR) chemical shifts, growth of resonances due to chemical exchange, and nuclear spin relaxation rates.^8–13 Temperature analysis via changes in the chemical shift are well-validated and highly adopted NMR-based techniques.^14–16 In MRI, temperature mapping can use the proton resonance frequency (PRF).^8,17 The PRF is a technique where the resonance frequencies of protons (¹H) from water are measured by applying a gradient-echo-based pulse sequence (dephasing and rephasing signals) and measuring the phase shift as a function of temperature to determine the phase coefficient. The difference in phase shifts can be made into a 2D temperature map to determine the relative changes in temperature. Recently, Gorny et al. were able to implement real-time PRF-based thermometry to accurately monitor microwave ablation treatments of hepatic tumors.¹⁸

The PRF method is commonly used due to the reported accuracy and maximum temperature responsiveness of 10 × 10⁻³ ppm °C⁻¹. However, discrepancies have been found in the PRF method depending on the type of gradient echo sequence used, which could introduce up to 28% error in temperature accuracy.¹⁰ Furthermore, the PRF method is less suitable for measurements of fatty tissues because the protons in adipose tissues (e.g., lipids) are less susceptible to temperature changes, making temperature changes in adipose tissues nearly undetectable.¹⁹ Additionally, the PRF is dependent on the magnetic field and is highly susceptible to magnetic inhomogeneity, where a magnetic field drift of 0.02 ppm h⁻¹ would generate an error of ±2 °C, during a 1 h measurement.¹¹

As complementary and alternative methods to ¹H MRI, fluorine-based NMR and MRI are gaining traction for preclinical and clinical diagnostic studies. The fluorine-19 (¹⁹F) isotope is 100% naturally abundant and has an NMR signal sensitivity for ¹⁹F that is 83% as sensitive as the ¹H signal.²⁰ Most importantly, for ¹⁹F MRI, there is no observable endogenous fluorine in the background of ¹⁹F MRI scans. Therefore, the localization of fluorinated probes is well defined and can be overlaid with ¹H MRI to correlate with an anatomical space. Perfluorocarbon (PFC)-based temperature sensors have been shown to measure temperature directly via their NMR parameters, for example, relaxation or chemical shift.^21–23 Relaxation is influenced by paramagnetic species (i.e., oxygen), which has been beneficial for dual oximetry–thermometry measurements but requires control of the oxygen and thermal environment for calibration. Alternatively, chemical shift thermometry, analogous to the use of ethylene glycol for NMR,^14,16 is independent of the oxygen environment. Currently, the state-of-art PFC-based thermometry sensor is perfluorotributylamine (PFTBA);²³ however, in terms of chemical shift thermometry, Berkowitz et al. reported a thermal responsiveness of ~9 × 10⁻³ ppm °C⁻¹, which limits its desirability, compared to PRF temperature mapping, despite its ability to directly determine absolute temperature.²⁴ Thus, there is a need for a more highly responsive and MR-based temperature probe to improve the accuracy of temperature measurements. With this goal in mind, we describe the rational design of a fluorine-based molecular temperature sensor with a nearly 2-fold improvement in temperature responsiveness as assessed by ¹⁹F NMR, compared to the PRF temperature mapping technique.

EXPERIMENTAL

¹⁹F NMR Variable Temperature Measurements.

Compounds were dissolved in CDCl₃ and measured on a Bruker AVANCE III 500 equipped with a 5 mm BBFO SmartProbe. Fluorine spectra were obtained at 471 MHz with dummy scans = 0, acquisition time = 2 s, delay time = 8.1 s, pre-scan delay = 100 μs, and the number of scans ≥ 8. The temperature was increased by passing heated N₂ gas over the spinning sample which was monitored by an internal instrument temperature probe; the temperature was allowed to stabilize for ≥ 5 min. At each temperature, the sample underwent shimming to correct for magnetic inhomogeneities. PFTBA measurements were made by either measuring neat PFTBA in an NMR tube with a flame-sealed capillary tube filled with D₂O, suspending PFTBA-filled mesoporous silica nanoparticles in H₂O/D₂O (9:1 v/v),²⁵ or by sonicating neat PFTBA in CDCl₃ for 10–15 min, then vortexing the sample prior to insertion into the instrument.

Computation of Temperature Sensors.

To find energetically favorable conformers at a given temperature, Monte Carlo searches were first carried out using the OPLS3 force field.²⁶ All identified minima were then re-optimized at the M06-L²⁷ density functional theory (DFT) level using the 6-31+G(d,p)²⁸ basis set. Harmonic vibrational frequencies were computed at this level to verify the nature of all stationary points and for use in molecular vibrational partition functions using the quasi-harmonic-oscillator approximation²⁹ (where all frequencies below 50 cm⁻¹ are replaced by values of 50 cm⁻¹). Single point DFT calculations at the M06-2X/6−311+G(d,p) level of theory were carried out for all optimized structures to obtain improved electronic energies to which thermal contributions to free energy were added from the M06-L/6−31+G(d,p) step.³⁰ To obtain the computationally predicted NMR spectra, single point DFT calculations on the minima were carried out at the B3LYP/6-311+G(d,p) level of theory,^31,32 and the gauge-independent atomic orbital method (GIAO) was used to obtain the chemical shifts.³³

DFT calculations were accomplished with the Gaussian 16 suite of electronic structure programs,³⁴ and force-field Monte Carlo searches were accomplished with MacroModel.³⁵

RESULTS AND DISCUSSION

PFTBA Temperature Sensing Is Enhanced in Chloroform.

The current state-of-art ¹⁹F NMR temperature sensor, PFTBA, is a commercially available compound with three fluorinated alkyl tails. Temperature measurements with PFTBA work by comparing the relative change in chemical shifts between two non-equivalent fluorine groups (e.g., CF₃ and γ-CF₂) on the fluorinated alkyl tail, where the NMR frequency difference between two fluorinated groups can be plotted as a function of temperature [i.e., ΔΔδ(T)] to determine responsiveness (Figure S1a–c). PFTBA was analyzed as a model sensor to benchmark temperature measurements in different solvents and chemical environments we expected to use for our new sensors (Figure 1). PFTBA was measured as either a neat liquid, within an aqueous environment encapsulated by previously reported mesoporous silica nanoparticles, perfluorocarbon-loaded ultraporous mesostructured silica nanoparticles (PERFUMNs),²⁵ or in CDCl₃. In the case of neat PFTBA, the α-CF₂ and β-CF₂ had the lowest thermal response of ~1 × 10⁻³ ppm °C⁻¹, while the γ-CF₂ and CF₃ were the most responsive combination with a thermal response of ~9 × 10⁻³ ppm °C⁻¹, which is consistent with the literature.²⁴

Figure 1.

Comparison of perfluorotributylamine (PFTBA) NMR thermal response as neat PFTBA, encapsulated as PERFUMNs in water, and in CDCl₃. The PFTBA structure is included on the right. Neat PFTBA has a thermal response similar to PFTBA PERFUMNs dispersed in water. Neat PFTBA dispersed in CDCl₃ has a significantly higher thermal response for the most sensitive pair of fluorine groups (positions γ and δ), indicating a solvent effect. There is no significant difference for the CF₂ groups near the center nitrogen (positions α and β). The p-value is <0.0001 for significant terms. Greek characters indicate the assignment of the fluorine group.³⁶

While there was no significant difference in thermal response between neat PFTBA and PFTBA-filled nanoparticles, the temperature response was significantly different for PFTBA in CDCl₃. Only the pairings that contained γ-CF₂ and CF₃ were different, demonstrating that the solvent effect is only significant to the most solvent exposed fluorine groups. Interestingly, the thermal response was diminished with γ-CF₂ pairings and enhanced with CF₃ parings, indicating opposing behavior between the fluorine groups. Thermal enhancement only raised responsiveness by 2–3 × 10⁻³ ppm °C⁻¹ per combination. In the case of the γ-CF₂/CF₃ pair, ΔΔδ(T) increased from 9.5 × 10⁻³ to 11 × 10⁻³ ppm °C⁻¹. Based on these findings, we sought to design and test fluorinated compounds with at least two or more magnetically distinct fluorine groups. Despite the small increase in responsiveness in CDCl₃, this solvent was chosen for further analysis due to its ability to dissolve a large range of fluorinated small molecules.

Computation Predicts Relative Temperature Sensing Performance between PFTBA and Commercial Fluorinated Compound.

In an effort to have a more directed rationale for the synthesis of novel compounds, we computationally determined the thermal response of several fluorinated compounds. The theoretical chemical shifts from computational analysis are based on the energetic contributions of conformers at a given temperature. These conformations are influenced by physiochemical phenomena such as steric strain, torsional strain, electrostatic interactions, and conjugation.^38,39 For appreciable temperature responses, two conditions must be met: first, the chemical shift of certain fluorine atoms must differ significantly across different conformers. Second, the energy difference between conformers with differing chemical shifts must be such that their populations vary substantially in the temperature range being measured. Decafluorobiphenyl was chosen as a test compound to compare to PFTBA because decafluorobiphenyl has three distinct fluorine groups (Figure 2a) directly attached to the biphenyl rings, which would limit the motion and number of conformations affecting the individual fluorine atoms, compared to a fluorinated alkyl tail.^40,41 The chemical shifts for PFTBA and decafluorobiphenyl were determined over a temperature range of 20 °C (Tables S1 and S2). Notably, the absolute value of the chemical shifts and temperature responsiveness were different from experimental values, where the best temperature responsiveness for PFTBA was 2.5 × 10⁻³ ppm °C⁻¹. The discrepancy in theoretical calculations is likely from differences in environmental and physicochemical conditions yet to be optimized (e.g., solvent, long-range interactions, state phases) and limitations from uncertainty in conformer energies, but comparisons between molecules are still useful.

Figure 2.

Temperature response of decafluorobiphenyl and comparison to perfluorotributylamine (PFTBA). (A) ¹⁹F NMR spectra of decafluorobiphenyl as the temperature increases. The assignments for fluorine groups are colored to match their assignment on the spectrum; (a) (blue) = 2,2′,6,6′-F, (b) (red) = 3,3′,5,5′-F, (c) (black) = 4,4′-F.³⁷ (B) Computational and (C) experimental comparison of thermal response [ΔΔδ(T)] between decafluorobiphenyl and perfluorotributylamine. The graph compares the quantified response of PFTBA vs decafluorobiphenyl based on the most and least responsive fluorine group pairs.

Despite an underestimation in theoretical responsiveness, compared to experimental, decafluorobiphenyl shows a much lower response to temperature compared to PFTBA, with PFTBA performing 2.5-fold (computational) and 2.9-fold (experimental) better than decafluorobiphenyl (Figure 2b,c). The difference in responsiveness between decafluorobiphenyl and PFTBA suggests that fluorine on a more dynamic alkyl tail would be more responsive than the fluorine groups appended to a ring. In consideration of the discrepancies, the relative trends in responsiveness for a library of compounds could still give information about where to focus for experimental synthesis. In consideration of the ~4-fold underestimation of PFTBA between theoretical and experimental values, we hypothesized that compounds with theoretical ΔΔδ(T) greater than 3 × 10⁻³ ppm °C⁻¹ would potentially overcome the 10 × 10⁻³ ppm °C⁻¹ PRF threshold.

Computation Suggests Best Enhancement from the Fluorinated Alkyl Tail Anchored onto a Conformationally Rigid Structure.

A library of theoretical compounds were tested with computational analysis over a temperature range of 20 °C and ordered based on their maximal responsiveness (Table 1). Compounds were initially selected based on their potential dynamic susceptibility to temperature, for example, the interconversion between restricted isomers on amide- and urea-based compounds causes conformational changes to be highly dependent on temperature.^42,43 We also hypothesized that an acetamide-based compound would lead to suitable temperature responsiveness if one fluorinated group was restricted and the other fluorinated group had a temperature-dependent conformational change.

Table 1.

Temperature Response of Theoretical Sensors

#	Structure	ΔΔδ(T) ppm °C−1	#	Structure	ΔΔδ(T) ppm °C−1
I		5.2 × 10−6	VII		8.1 × 10−4
II		3.0 × 10−5	VIII		9.4 × 10−4
III		2.0 × 10−4	IX		1.8 × 10−3
IV		2.2 × 10−4	X		2.1 × 10−3
V		2.3 × 10−4	XI		2.9 × 10−3
VI		5.4 × 10−4	XII		11 × 10−3

Despite our predictions, most of the amide- and urea-containing compounds had poor temperature responsiveness (≪ 1 × 10⁻³ ppm °C⁻¹). Comparing I and II, the change from two meta-CF₃ to one para-CF₃ on the phenyl ring resulted in a 6-fold increase to the thermal responsiveness. At any given temperature, the weighted dispersion of chemical shifts predicts that the acetyl-CF₃ groups have the largest spread of chemical shift populations (Table S3), suggesting inherent responsiveness with certain functional groups. For both I and II, acetyl-CF₃ was the most responsive fluorine group (Table S4). A 38-fold increase in ΔΔδ(T) occurred (relative to I) when converting to (thio)urea- and carbamate-based compounds (III–V). Compounds III–V also show slight changes in temperature responses when different heteroatoms are incorporated at the core of the urea-based structure. Furthermore, the effects of the amide bond rotation appeared to be minimal because only a slight increase in the thermal response was seen between III and IV when the urea nitrogen was changed to an oxygen atom in the carbamate. Alternatively, compound VI had over a 2-fold increase from III–V, suggesting that using a non-benzylic CF₃ group may be beneficial. Interestingly, VII still had a 1.5-fold enhancement despite going from urea to acetamide. Having less CF₃ groups and/or a substituent in the para-position enhanced the theoretical temperature response with VII, as seen with I to II. Of note, with benzylic CF₃ groups, the fluorine atoms are symmetrically correlated such that their degenerate conformations, which are identical in energy and geometry but not labeling, may not be captured (Figure S2). With only specific ordering of fluorine atoms captured, the predicted thermal sensitivities are artificially increased. Averaging the chemical shift of these equivalent fluorine accounts for this symmetry.

Among the initial computational results (I–VII), we only considered trifluoromethyl groups. On these analyses, it was apparent that these substituents on the phenyl ring were insufficient to enhance thermal responsiveness. Greater improvement was seen when fluorinated alkyl tails were incorporated into the chemical structures (VIII–XII). VIII was analyzed to benchmark effects of a perfluoropropyl-alkyl tail. Compounds IX–XII approached our theoretical threshold with ΔΔδ(T) ranging from 1.8–2.9 × 10⁻³ ppm °C⁻¹ thermal responses. Incorporation of two different fluorinated alkyl chains on the phenyl rings of IX led to a 3-fold increase in responsiveness relative to VI, the second most responsive urea. On X, we combined the phenyl CF₃ with a fluorinated alkyl tail. The most responsive pairing was between the phenyl CF₃ and the alkyl CF₂ groups, suggesting that a fluorine group on a rigid structure (e.g., phenyl ring) in combination with a fluorinated alkyl tail anchored to the rigid structure would greatly increase responsiveness. Of interest, the benzylic CF₃ had the same degenerate correlated conformers, as seen in V, but the alkyl CF₃ appeared to have three distinct groups of conformers (Figure S3).

While compound X showed promising results by mixing functional groups on rings and fluorinated alkyl tails, results from compounds XI and XII suggest that good enhancement can be achieved by simply anchoring the fluorinated alkyl tails onto a conformationally rigid structure. Compounds XI and XII were generated by appending the fluorinated tail of VIII onto a hydroquinone or diazapyridinophane core, respectively. Compound XII had the strongest thermal responsiveness with ΔΔδ(T) = 11 × 10⁻³ ppm °C⁻¹. Therefore, given an underestimation of the theoretical ΔΔδ(T), these computational analyses supported appending a fluorinated alkyl tail on a larger conformationally rigid core structure to enhance the experimental responsiveness.

Hybrid Compounds with Fluorine Groups on a Rigid Structure and Fluorinated Alkyl Tails Are Necessary to Enhance Effects of the Fluorinated Alkyl Tail Alone.

As experimental validation for the pattern of responsiveness found in the computational library, several compounds were synthesized to confirm the different effects of structural features on temperature responsiveness, and the theoretical thermal response of these new compounds were also computed (Table 2). The computational values for the thermal response correlate with the experimental values, except for compound 4, which is an outlier. Compounds 1 and 2 were synthesized with the expectation that the rotational barrier of the thiourea moiety would hinder the rotation of the R groups around the two nitrogens (Figure S4a,b). As expected, both compounds performed poorly with ΔΔδ(T) of 0.40 × 10⁻³ and 2.5 × 10⁻³ ppm °C⁻¹, respectively (Figure S4c,d). Based on these results, we moved forward to measure the effects of a fluorinated alkyl tail attached to a rigid structure.

Table 2.

Thermal Response of Synthesized Compounds^a

#	Structure	ΔΔδ(T), ×10−3 ppm °C−1 (Experimental)	ΔΔδ(T), ×10−3 ppm °C−1 (Theory)
1		0.4	0.23
2		2.5	0.30
3		10.1	0.92
4		10.4	0.24
5		12.5	2.06
6		19.5	5.92a
7		18.9	7.21a
XIII		ND	8.24

^aTo reduce the number of conformations and computational time, a model compound was used with a shorter fluorinated tail. The structures can be found in the Supporting Information.

A simple fluorinated tail was needed to test the hypothesis derived from the computational library. Therefore, 2,2,3,3,3-pentafluoropropylamine was selected due to its commercial availability, two distinct fluorine groups (CF₂ and CF₃), and a methylene between the alkyl fluorine groups and the nitrogen to allow for better reactivity for synthesis. With the pentafluoropropylamine as a precursor, compounds 3 and 4 (Figures S5a, S6a) were synthesized via acylation reactions to append a fluorinated tail onto a rigid structure. Encouragingly, 3 and 4 had a responsiveness at the PRF threshold with ΔΔδ(T) = 10.3 × 10⁻³ and 10.4 × 10⁻³ ppm °C⁻¹, respectively (Figures S5b, S6b), which is also comparable to PFTBA measured under these conditions (11 × 10⁻³ ppm °C⁻¹). The responses from 3 and 4 support the hypothesis that the fluorinated alkyl tail on a rigid structure would have a strong temperature response. In addition, the CF₂ group was the most responsive to temperature changes. The CF₂ on 3 had an individual responsiveness [i.e., Δδ(T)] of 8.4 × 10⁻³ ppm °C⁻¹; for CF₂ on 4 Δδ(T) = 8.8 × 10⁻³ ppm °C⁻¹ (Figures S5c, S6c). Interestingly, the resonances of the CF₃ groups on 3 and 4 changed as a function of temperature in the opposite direction relative to the CF₂ group, with lower magnitudes of −1.7 × 10⁻³ and −1.6 × 10⁻³ ppm °C⁻¹, respectively, where the negative term was given to upfield movement on the NMR scale.

Compounds 3 and 4 were also made to test if the space between the fluorinated alkyl tail and the rigid structure significantly contributed to the thermal response. While both compounds were able to overcome the PRF threshold, the distance from the rigid body by incorporation of a methylene (4) or ethylene (not shown) did not significantly change the ΔΔδ(T) between compounds (Table S5). Surprisingly, further investigation revealed ΔΔδ(T) = 10.5 × 10⁻³ ppm °C⁻¹ for the precursor 2,2,3,3,3-pentafluoroproplyamine, indicating that the overall responsiveness was already around its limit and appending the fluorinated tail onto a rigid structure did not enhance the endogenous temperature response of the precursor (Table S6). Therefore, we hypothesized that a hybrid compound that incorporates fluorine groups on a rigid core with the anchored alkyl tail could experimentally overcome a desired 1.5-fold enhancement relative to the PRF threshold.

Because the theoretical data in Table 1 suggested that compound X is a promising hybrid compound, compound 5 was synthesized with the incorporation of benzylic CF₃ groups as the fluorine group on the rigid core, while also maintaining the response of the CF₂ on the fluorinated tail (Figure S7a). Consistent with our design hypothesis, 5 improved ΔΔδ(T) to 12.5 × 10⁻³ ppm °C⁻¹, making 5 approximately 1.3-fold more responsive than the PRF threshold (Figure S7b). Interestingly, the weighted dispersion via computation identified the CF₂ as the most responsive group and even suggested the correct pairing for the maximal temperature response (Figures S8 and S9). Individually, the movement of the benzylic CF₃ and alkyl CF₂ resonances was similar in magnitude but opposing in direction to temperature changes, as seen with 3 and 4. Overall, Δδ(T) were −5.9 × 10⁻³ and +6.7 × 10⁻³ ppm °C⁻¹ for the benzylic CF₃ and alkyl CF₂ groups, respectively (Figure S7c). Given the opposing behavior of these functional groups in response to temperature changes, we sought to model this relationship, where temperature sensitivity is enhanced due to varying upfield/downfield frequency changes rather than having an insensitive–sensitive fluorine group pairing. Based on these relationships, we were able to formulate a new screening method to match the fluorine groups to enhance overall thermal sensitivity.

Screening Method to Match Different Fluorine Groups Together to Make Novel Sensors.

The mathematical relationship between ΔΔδ(T) and the individual movements of fluorine groups is linear (Supporting Information “Linear Model for Screening”). As chemical shifts of two fluorine groups move as a function of temperature, at any given temperature, the magnitude between groups is equal to the difference in frequency, that is, ΔΔδ. Thus, the responsiveness, ΔΔδ(T), is equal to the magnitude between the slopes of the two functional groups (Figure 3a,b). Based on the relationship determined in our linear model, the more positive (downfield) and negative (upfield) two resonances become, as a function of temperature, the more the magnitude of ΔΔδ(T) will increase. Experimental data were compiled onto a scale to highlight the relationship between the movements of different fluorine groups (Figure 3c). Assuming that most fluorine groups, as seen experimentally, are not inherently responsive beyond 10 × 10⁻³ ppm °C⁻¹, the PRF threshold was used as the boundary for the majority of fluorine groups’ responsiveness to temperature. The propylamine tail (orange) shows the CF₂ group is beyond the threshold, while the CF₃ group is between no response and the midpoint.

Figure 3.

Computational screening method of functional groups to develop novel fluorinated sensors. (A) Theoretical response of two different fluorine resonances (Δδ_A and Δδ_B) as a function of temperature in a linear model. At a given temperature, the difference between Δδ_A and Δδ_B is equal to the ΔΔδ. (B) Theoretical graph of how ΔΔδ(T) is based on the absolute difference between the slopes of the two functional groups (Δδ_A and Δδ_B) as a function of temperature [ΔΔδ(T) = m_D]. (C) Scale and boundary of experimental fluorine group responsiveness. The ideal case is to identify two functional groups in the more common regions of responsiveness (e.g., ±7 × 10⁻³ ppm °C⁻¹) with opposing directions in response to temperature. The absolute magnitude of difference between those two responsive groups is the overall responsiveness of the temperature probe and should be ≥1.5 times the PRF signal. The values under each molecule are from experimental measurements. Color coded numbers are different functional groups on the same molecule. Each fluorine group on PFTBA (right side) has an individual response from 18 to 27 × 10⁻³ ppm °C⁻¹.

Essentially, the closer CF₃ Δδ(T) (i.e., m_A or m_B; Figure 3) is to the CF₂ group the more the magnitude of ΔΔδ(T) reduces for the molecule. Similarly, the effect can be seen with the tail of PFTBA (gray) and in the opposing direction (negative) with compound 2 (green), where Δδ(T) of the fluorine groups are both similar in magnitude and direction. Thus, while most Δδ(T) are within the ±10 × 10⁻³ ppm °C⁻¹ boundary, the PRF threshold can be readily overcome by finding a pair with Δδ(T) ≤ −5 × 10⁻³ and Δδ(T) ≥ +5 × 10⁻³ ppm °C⁻¹. Furthermore, if the weighted dispersion can correctly identify the most and least responsive fluorine groups, as seen with 5, this screening model becomes a tool that can be used with theoretical computation. The modularity from analysis of fluorinated motifs versus fluorinated molecules is envisioned to drastically reduce the chemical space necessary to identify suitable temperature sensors.

Fluorinated Alkyl Tail, Rigid Core Aryl Fluorine, and a Thioether Significantly Enhance Thermal Responsiveness over the PRF Threshold.

We sought to test the screening results suggesting the combination of two different fluorinated motifs to generate novel temperature sensors by leveraging synthesis and computational analysis of a new hybrid compound. In general, the phenomenon where one resonance moves downfield while the other moves upfield is likely related to molecular electronics, as well as conformational changes. Previously, analysis with compounds III–V and experimental data with precursors for our temperature sensors (Table S6) showed that heteroatoms also impact the thermal response. These differences could be due to a difference in inductive or electron withdrawing effects, hydrogen bonding (if allowed), and/or strain on conformational transition states due to lone pair electronic and valence geometries.

Given the response of the aryl fluorines in our screening model, a perfluorobiphenyl-based test compound was chosen for computational analysis (Figure S10a). Experimental temperature responses revealed that the majority of the fluorines on both decafluorobiphenyl and perfluoro-4,4′-biphenol are within the desired negative region for temperature response, that is, −5 to −10 × 10⁻¹ ppm °C⁻¹ (Figure S10b,c). Additionally, the aryl fluorines, meta- and ortho- to the 4,4′-position, changed their responsiveness when the 4,4′-fluorine was changed to 4,4′-diols, demonstrating a significant effect from the heteroatom in the 4,4′ position. Three perfluorobiphenyl derivatives were thus computationally analyzed to simultaneously screen for a potentially improved temperature sensor and investigate the heteroatom effect on responsiveness between nitrogen, oxygen, and sulfur (Figure S11). The analysis revealed that sulfur (XIII, Table 2) had the best response among the three derivatives tested. The computation also supported the idea that the resonances of the biphenyl fluorines will not move similarly to the fluorine on the alkyl tail, possibly leading to responses in different directions. No strong correlation was found between the relative orientation of the fluorine atoms to sulfur and their chemical shifts. However, the weighted dispersion indicated that the first CF₂ would be the most responsive fluorine group and would pair with an aryl-fluorine resonance (Table S7).

Given the results from the computational study, compound 6 (Figure 4a) was synthesized with a heptadecafluorodecane-1-thiol (due to commercial availability). For the heptadecafluorodecane-1-thiol precursor, ΔΔδ(T) = 13.3 × 10⁻³ ppm °C⁻¹ (Table S8). Compound 6 had a significant increase in ΔΔδ(T) of 19.5 × 10⁻³ ± 0.2 × 10⁻³ ppm °C⁻¹ (n = 3), corresponding to ~2-fold increase in temperature responsiveness relative to the PRF, 1.8-fold increase from PFTBA in CDCl₃, and a 1.6-fold increase from 5 (Figure 4b–d). Additionally, 6 is made by substituting the alkyl tail onto the biphenyl ring in the 4 and 4′ positions. This substitution pattern maintains an axis of symmetry leading to a doubling of the magnetically equivalent fluorine atoms for better signal-to-noise ratios. Most of the resonances had Δδ(T) in the hypothetically optimal region (±5 to ±10 × 10⁻³ ppm °C⁻¹) indicating that almost all combinations could overcome the PRF threshold. The maximum response (green triangles) contained the aryl fluorines, meta to the thioethers (−136 ppm), with Δδ(T) = −7 × 10⁻³ ppm °C⁻¹, and the first CF₂ groups on the alkyl tail (−113 ppm), adjacent to the CH₂, with Δδ(T) = +12 × 10⁻³ ppm °C⁻¹ (Figure 4e). These results support the hypotheses of the screening method, where (A) the majority of the resonances were within the boundary; (B) pairs of resonances that were past ±5 × 10⁻³ ppm °C⁻¹ were able to easily overcome the PRF threshold; and (C) the combination of aryl fluorine with a fluorinated alkyl tail significantly improved the overall temperature response. In addition, the data suggest that for these partially fluorinated alkyl tails, the CF₂ adjacent to CH₂ will have the largest shift in resonance during changes in the temperature as consistent with the weighted dispersion calculation. While the combination of a thioether, fluorinated alkyl tail, and aryl fluorines significantly improved the responsiveness of our lead compound, we did not know if the presence of the biphenyl rings also had a significant impact on the temperature response. Therefore, compound 7 was synthesized, in similar fashion to 6, with a hexafluorobenzene for the rigid structure (Figure S12a).

Figure 4.

Structure and temperature response of 6. (A) Synthesis and structure of 6. (B) ¹⁹F NMR spectrum of 6 in CDCl₃. Assignments were derived from analogous structures in the literature.^37,44,45 (C) ¹⁹F spectra of most temperature responsive chemical shifts (ppm). (D) Temperature response curve of 6 with ΔΔδ(T) = 19.5 × 10⁻³ ± 0.2 × 10⁻³ ppm °C⁻¹ (n = 3, mean ± SEM). Compound 6 has a ~2-fold increase in responsiveness compared to the PRF. (E) Movement of individual resonances over 20 °C; several of the shifts are in the optimal region (±5 to ±10 × 10⁻³ ppm °C⁻¹) indicating most of the resonance combinations are above 10 × 10⁻³ ppm °C⁻¹. The most responsive combinations are indicated as green triangles (down for negative movement and up for positive movement). Error bars are negligible from three replicate measurements.

The most responsive pair of chemical shifts for compound 7 (first CF₂ and aryl fluorine) had a comparable ΔΔδ(T) of 18.9 × 10⁻³ ± 0.4 × 10⁻³ ppm °C⁻¹ (n = 2); there was no statistically significant difference between the thermal response of 6 and 7 (Figure S12b,c). While there is no significant difference, compound 6 does have a higher responsiveness indicating that the presence of the biphenyl rings has a minor effect on the thermal response. The biphenyl ring, compared to a benzene, should have more conformations and therefore more discrete orientations for the substituents to occupy. A more thorough investigation into the rigid body will be needed to understand how fused rings, linked rings, and different size rings can affect the temperature response of these organofluorine temperature sensors.

The long-term goal of this work is to incorporate the temperature-sensitive fluorocarbons into ultraporous mesostructured silica nanoparticles (UMNs).^25,46 Both compound 6 and 7 were white solids after purification, which limits their direct use as sensors, given the tendency for spin–spin relaxations of solids to lead to a significant line-broadening of NMR spectra.⁴⁷ Solid PFCs are difficult to load into UMNs and our efforts to solubilize these molecules with surfactants have proven challenging; however, diffuse loading and further sequestering solid particles inside the nanoparticles are plausible for use in future investigations.^48–50 Additionally, derivative designs, by shortening the fluorinated tail, or using different rigid structures may lead to a fluorocarbon liquid. Given the 1.3-fold increase in responsiveness with 5, and the increase in theoretical responsiveness between 6 and XIII, shorter tails are anticipated to be sufficient to enhance thermal sensitivity of analog compounds. While the heptadecafluorodecane-1-thiol was used based on commercial availability, shorter fluorinated alkyl thiols, such as those needed for making XIII are also anticipated to lead to reduced bio- and environmental accumulation which will be the focus of future work.

CONCLUSIONS

This work describes the rational design of temperature-responsive fluorinated molecules to act as temperature sensors using NMR. To address the large chemical space challenge, the synthetic possibilities were reduced and focused by incorporation of experimentally analyzed sensors and computationally driven hypotheses of the physiochemical properties that are crucial to the development of these fluorine-based temperature sensors. The best temperature sensor, 6, was discovered after supporting several hypotheses. The theory behind a “good” sensor is based on the following guidelines discovered in this work: a good sensor has (1) at least two chemically distinct fluorine groups on the same molecule; (2) a fluorinated alkyl tail and fluorine on a chemically rigid structure; (3) at least two fluorine groups with opposing upfield/downfield movement; and (4) the the opposing fluorine groups have individual thermal responses past ± 5 × 10⁻³ ppm °C⁻¹. Further investigation on the effects of heteroatoms are warranted and may improve the response of the sensors further. Additionally, investigation into different rigid chemical structures may improve thermal responsiveness while shorter fluoroalkyl tails may alter the physicochemical properties of the molecules to facilitate encapsulation into nanoparticles.

While measurements were done in CDCl₃, a nearly 2-fold increase in responsiveness to temperature based on the state-of-the-art MRI probes makes 6, and analogs thereof, new candidates for MRI temperature mapping and subsequent in vivo studies, either incorporated into UMNs or a different biological vehicle (e.g., liposomes, emulsions). The capability of taking more accurate and sensitive temperature measurements will lead to a better understanding of chemical and biological interactions, improved monitoring of heating systems, and more reliable diagnostics for evaluation of medical interventions.