Raman spectroscopy enabled investigation of carbon nanotubes quality upon dispersion in aqueous environments

Carbon nanotubes (CNTs) are all-carbon tubular structured materials with diameters in the range of 0.4–100 nm and lengths up to several micrometers.¹ They are considered ideal candidates for numerous biological and biomedical applications^2–5 owing to their extraordinary chemical^6,7 and physical properties.^8,9 Specifically, research has demonstrated the great promise of CNT-based polymer composites as electrochemical biosensors,^10,11 drug delivery agents,^12,13 gene transfection vectors,¹⁴ and biological imaging mediators.¹⁵ 

While pristine and in situ grown CNTs could be directly integrated into many inorganic systems, a highly stable suspension of dispersed CNTs in aqueous solutions is often a preliminary requirement for their use in biological and biomedical applications.¹⁶ However, not only CNTs are insoluble in aqueous solutions but also they tend to form aggregates due to substantial van der Waals interactions between the individual tubes.¹⁷ Recently, significant advances in the chemical modification,^18,19 biological functionalization,²⁰ as well as bioactive species conjugation^21,22 of CNTs have enabled researchers to obtain stably dispersed CNT aqueous solutions and, hence, to develop functional CNT-composite materials. It is important to note that these surface modification techniques not only lead to stably dispersed CNTs, they are often necessary for preparing biologically compatible CNT surfaces.¹⁴ 

Further, it has been shown that the adsorption of polymers and surfactants in conjunction with sonication, could greatly improve the dispersion of CNTs in aqueous solutions.^21–25 However, experimental observations as well as theoretical calculations have shown that dispersion methods supplying high energy input (such as sonication) can also induce CNTs' fracture and fragmentation,^26,27 thereby potentially diminishing the quality of CNTs during sonication. Despite its significance to the applications of CNT-based composites, this trade-off between the quality and dispersion of CNTs has not been studied systematically and part of the reason lies in the unavailability of instrumentation that enables such investigations.

Conventionally, SEM and TEM imaging techniques are often used to assess CNT dispersions because of their ability to provide direct visualization of CNT distribution within composites materials.^28,29 However, they provide only a qualitative understanding of structure, rather than a quantitative understanding of quality and require extensive pretreatment steps that could damage the samples. As a premium nondestructive characterization tool, Raman spectroscopy could provide valuable structural information about samples, while preserving their integrity.³⁰ Furthermore, Raman spectroscopy of carbon nanomaterials including CNTs have been studied extensively and Raman spectra of various polymers are also well-known,^31–34 making it a suitable candidate to study the structure and quality preservation of CNTs post-sonication.

In this report, we investigate how different experimental parameters affect the quality of CNTs during dispersion using Raman spectroscopy, in order to enhance our understanding of the relationship between CNTs' quality and dispersion parameters. We focused on multi-walled CNT (MWCNT)-polyethylene glycol (PEG) solutions. PEG was chosen due to its hydrophilicity, biocompatibility, inertness, and tunable physical, mechanical, and biochemical properties closely emulating most soft tissues in the body. PEG has been approved by the U.S. Food and Drug Administration for use in biomedical devices, and PEG-CNT composites, in particular, have recently gained interest in various biomedical applications, such as stem cell scaffolds, cell delivery vehicles, and electrode coatings.^35–38 

Specifically, by analyzing the Raman spectra of MWCNT-PEG solutions, we established correlations between MWCNTs' quality and dispersion as a function of surfactants type and concentration, sonication time, sonication method, and MWCNTs' geometry. Our results could serve as a roadmap for the development of optimal experimental setup to control CNTs' quality during dispersion in polymeric solutions and, therefore, enable the development of effective CNT-polymer composites for targeted biological applications. Furthermore, because CNT modifications for biological applications are mainly determined by the surface properties of the CNTs and the amount of defects present, this study could also aid further understanding of CNT functionalization. For example, the entire surface of CNTs can be modified via chemisorption or physisorption by depositing a specific molecule of choice. Additionally, changes to carbon surfaces can be done via covalent bonding, which acts by disrupting the intrinsic structure of the carbon lattice and results in vacancies or defects. All of the above described changes can be detected in Raman spectral plots, including monitoring the evolution of those changes with time.

All reagents were purchased from Sigma Aldrich (St. Louis, MO) and used as is unless otherwise noted. Other materials used in this study were obtained as follows: PEG-diOH (MW 3365 Da) from Electron Microscopy Sciences (Hatfield, PA), bovine serum albumin (BSA), and dimethyl sulfoxide (DMSO) from Fisher Scientific (Fair Lawn, NJ), 700 nm-thick thermal evaporation prepared silicon/silicon dioxide (Si/SiO₂) wafers from WRS materials (Spring City, PA), KimWipes from Kimberly-Clark Professional (Roswell, GA), and isopropanol from VWR (Wester Chester, PA). Two samples of MWCNTs, MRCMW10 and MRCMW70, produced via catalytic chemical vapor deposition were generously provided from MerCorp (Tucson, AZ). The dimensions for MRCMW10 (81% purity; termed long) were 20 ± 3 nm outer diameter and 3 ± 2 μm length, while the dimensions for MRCMW70 (>90% purity; termed short) were 10 ± 1 nm outer diameter and ∼1.5 pm in length.

To prepare MWCNT-PEG solutions, first, 20% w/v stock solution of PEG was prepared by dissolution in 0.3 M triethanolamine (TEA) in phosphate buffered saline (1×, and pH 8). To prepare a working MWCNT-PEG solution, 0.05% w/v MWCNTs were mixed with PEG and additional TEA buffer was added to achieve a final PEG concentration of 10% w/v. Then, the MWCNTs were dispersed in the solution using a probe sonicator (Fisher Scientific, Waltham, MA) operating at a frequency of 4 kHz for varying times. Alternatively, a continuous ultrasonic water bath sonicator (Branson Ultrasonics Corp, Danbury, CT) was used for 5 min at an output of 110 W and 40 kHz. Additionally, BSA or DMSO were added as surfactants in some cases to aid the dispersion. All experiments are described in Table I in detail and schematically represented in Fig. 1. To investigate the effect of sonication time as well as MWCNTs' geometry on the quality and dispersion of MWCNTs, we dispersed two types of MWCNTs, namely, MRCMW10 (termed long) and MRCMW70 (termed short), in PEG solution containing 0.05% w/v BSA for 0, 5, 10, and 20 min. To investigate the effect of surfactants, we dispersed long MWCNTs (MRCMW10) in PEG solutions containing 0.05% w/v BSA or 1% w/v DMSO, each for 10 min. Finally, to investigate the effect of sonication conditions on MWCNTs' quality, we dispersed long MWCNTs (MRCMW10) in PEG solutions containing 0.05% w/v BSA for 5 min using either water bath sonication or probe sonication.

Upon dispersion, phase contrast images of the solutions were taken at 10× magnification (Zeiss, Axiovert 200M, Oberkochen, Germany). For absorbance measurements, dispersed MWCNT-PEG solutions were placed in a quartz cuvette and absorbance (300–900 nm scan) was measured on Spectra MAX Plus spectrophotometer (Molecular Devices, Sunnyvale, CA). Absorbance values of surfactant solutions in the absence of MWCNTs were taken as a baseline and subtracted from the absorbance readings of the solution in the presence of MWCNTs. Better dispersion was correlated to increased absorbance values.²³ The dispersed MWCNT-PEG solutions were then pipetted onto Si/SiO₂ chips and left to dehydrate at 60 °C for 24 h. The Si/SiO₂ chips were cut (∼1 × 1 cm²) and cleaned by sonication in sequence of deionized-water (EMD Millipore Corp, Darmstadt, Germany), acetone, and finally isopropanol, and dried with nitrogen gas. Dried MWCNT-PEG solutions on Si/SiO₂ chips were then characterized using Raman spectrometer (Renishaw InVia, Hoffman Estates, Illinois) with a 532-nm laser. The laser was focused onto the samples with a 0.75 NA/50× objective; hence, the laser spot diameter at the probed area of the sample was estimated conservatively to be in the range of 1.1–1.2 μm. The expression to determine the minimum possible spot size of Raman measurement is as follows:

where NA is numerical aperture and λ is the laser excitation wavelength. All Raman spectral plots were acquired in the range of 1100–3200 cm⁻¹, capturing the intervals that contain main characteristic Raman bands of MWCNTs as well as PEG.

Numerical results are reported as averages ± standard deviation. Statistical significance between multiple samples was tested by analysis of variance followed by a posthoc analysis and between two samples by a Student's t-test (p < 0.05). A minimum of three samples from three independent experiments were tested per condition.

Raman spectroscopy studies the inelastic scattering of light in various materials. In a typical resonant Raman scattering process, an incident photon from a laser source is absorbed by an electron and the electron is promoted from the valence band (VB) to the conduction band, leaving a hole behind in the VB. The electron later recombines with the hole through one or more scattering events such as emitting (Stokes) or absorbing (Anti-Stokes) phonons.³² Therefore, Raman spectroscopy could provide much information about the electronic and phonon (vibrational) structures of various materials. Specifically, researchers have been able to extract structural information such as diameter, chirality and purity from Raman spectra of CNTs.³¹ For MWCNTs, there are three main characteristic Raman bands. The G band, located at ∼1580 cm⁻¹, is the result of vibrations of the carbon atoms along the graphite walls of MWCNTs. The D band, located at ∼1350 cm⁻¹, is a product of second order, two phonon processes and can represent the amount of disorder (impurities) present in the carbon crystal structure as well as other types of structural damage such as fragmentation or deformation. The G′ band, appearing at ∼2700 cm⁻¹, is the second harmonic of the D band.³³ Importantly, the ratio of the intensity of the G band to the intensity of the D band is commonly referred to as the quality factor of CNTs and can be expressed as

Since the presence of CNTs' crystal structure contributes to the G band intensity, and the presence of disorder in CNTs' structure contributes to the D band intensity, higher Q_CNT corresponds to better CNTs' quality³⁹ On the other hand, the PEG solutions used in our study have two main characteristic Raman bands situated at 1500 and 2800–3100 cm⁻¹ representing the vibrations of various chemical bonds.³⁴ 

Figure 2 shows representative Raman spectra of MWCNTs (a), a PEG solution (b), and a MWCNTs-PEG solution (c). From (a) and (b), we can clearly identify the characteristic Raman bands of both MWCNTs (D, G and G′ as labeled) and PEG solutions as described above. Moreover, by comparing (a) and (b) with (c), we see that after mixing, well-defined and distinguishable characteristic Raman bands for both materials appeared at the same locations (without shifts) on the Raman spectral plots as before mixing. This indicates that the interaction between PEG and CNTs' surfaces did not influence the structural properties of CNTs, which would otherwise be detected in the Raman signal. Hence, Raman spectroscopy provides a convenient and direct way to examine the quality of CNTs [refer to Eq. (1)].

While not addressed here, Raman spectroscopy can also be applied to analyze single-walled carbon nanotubes (SWCNTs) quality factor, Q_SWCNT, in aqueous environments before and after mixing in a similar fashion as it is done for MWCNTs.^40,41 Specifically, in a Raman spectrum of SWCNTs, the characteristic Raman bands (D, G, and G^′) would also be present.^31,32 The G band in carbon nanotubes occurring in the 1500–1605 cm⁻¹ range is an intrinsic feature that is closely related to vibrations in all sp² carbons.

The difference, however, would be that while MWCNTs generally exhibit metallic properties, SWCNTs can be either metallic or semiconducting and their spectral characteristics, particularly the G band, will differ. Specifically, the G band in semiconducting SWCNTs usually shows two dominant features described as G⁻ and G⁺ peaks with Lorentzian line shapes and narrow width (full width at half maximum, ∼6–15 cm⁻¹). The lower frequency component is associated with vibrations along the circumferential direction (G⁻), and the higher frequency component (G⁺) is attributed to vibrations along the direction of the nanotube axis.³³ In metallic SWCNTs Raman spectra will also have two dominant components with similar origins; however, the G⁺ has a Lorentzian line shape as that for the semiconducting nanotubes, but the G⁻ has a very broad Breit–Wigner–Fano line shape. For measurements on the quality of SWCNTs one would generally consider sp² characteristic feature at ∼1580 cm⁻², which represents vibrations along the nanotube axis, so Q_SWCNT can be determined as $IG+$/I_d for either semiconducting or metallic SWCNTs. Thus, the evaluation of samples containing SWCNTs can be performed in the same fashion as for MWCNTs with some variations in the sample preparation. As indicated earlier, because SWCNTs often contain a mixture of semiconducting and metallic CNTs, those might need to be separated prior to Raman measurements. Techniques to selectively isolate SWCNTs of specific diameter and electronic properties have been well-established.^42–44 

Ultrasonic energy is used to achieve MWCNTs dispersion through the following three-step cycle: (1) ultrasonic energy delivered to the solution facilitates the formation of cavitation bubbles at the surfaces of MWCNTs, (2) cavitation bubbles rapidly expand and push MWCNTs within a bundle apart, and (3) bubbles violently implode after reaching a size threshold and further disperse the MWCNTs. During steps (2) and (3), much stress is being generated at the walls of MWCNTs causing tube fracture, deformation, and fragmentation.²⁶ As shown in the schematic diagram in Fig. 1, starting with insoluble MWCNTs in an aqueous environment (left), we added surfactant (0.05% w/v BSA) and polymer (10% w/v PEG) to the solution and then used sonication to obtain dispersed MWCNTs (right). We then used Raman spectroscopy, specifically Q_cnt, to quantify the change in MWCNTs quality caused by ultrasonic energy. We also used absorbance spectroscopy and optical imaging to concurrently evaluate the dispersion of MWCNTs.

Figure 3(a) shows the Raman spectra of MWCNT-PEG solutions after 10 min of sonication in BSA (top panel) and DMSO (bottom panel). The main Raman bands of both MWCNTs and PEG were clearly distinguishable and average values of Q_CNT are shown. For MWCNTs sonicated in the presence of BSA, Q_CNT was 0.9337 ± 0.0059, whereas Q_CNT of MWCNTs sonicated in DMSO solution was 0.9007 ± 0.0045. Hence, after 10 min of sonication, Q_Cnt of MWCNTs dispersed in BSA was significantly higher than that of MWCNTs dispersed in DMSO (p < 0.01). Both values were significantly lower than Q_CNT of non-sonicated MWCNTs (Q_CNT = 1.0187 ± 0.0134; see Table I) as anticipated (p < 0.005 for BSA and p < 0.005 for DMSO). Compared to no-sonication control, the decrease in Q_Cnt was 8.3% when BSA was used as a surfactant, while it was 11.6% when DMSO was used. Figure 3(b) shows the corresponding absorbance spectra (for 10 min sonication) for MWCNT-PEG solution containing BSA (solid line) or DMSO (dashed line). The absorbance of the MWCNT-PEG solution containing BSA was higher than the one containing DMSO, indicating a more uniformly black solution and, hence, a better dispersion. Figure 3(c) shows optical images of the MWCNT-PEG solutions, where MWCNTs aggregates were visible as black clusters. As with the absorbance data, we observed that the solution containing BSA (top panel) was more dispersed (i.e., smaller MWCNTs aggregates) than the solution containing DMSO (bottom panel).

Our results indicated that not only was BSA a better dispersion agent than DMSO, it also appeared to better protect MWCNTs from mechanical disruption induced by ultrasonication. It has been shown previously that because BSA contains both hydrophobic and hydrophilic residues, when dispersed in MWCNTs solutions, the hydrophobic residues interact with MWCNTs while the hydrophilic residues interact with water, resulting in stable MWCNTs aqueous dispersions.⁴⁵ To verify the possible MWCNTs protection effect from surfactants observed in our data, we performed additional control experiments on surfactants and polymers. Namely, we dispersed MWCNTs in aqueous solutions with and without surfactants (BSA, DMSO) or polymers (PEG). The detailed experimental parameters and the resulting values of Q_CNT are shown in Table I. We noted that compared to MWCNTs dispersed in just water (Q_CNT = 0.8462 ± 0.0301), MWCNTs dispersed in water containing DMSO, BSA and PEG all resulted in higher Q_CNT values of 0.8665 ± 0.0428, 0.8848 ± 0.0242, and 0.9396 ± 0.0683, respectively. Among all additives, PEG had the most pronounced effect on MWCNTs quality preservation.

To study the effect of sonication time on MWCNTs' quality during dispersion, we dispersed MWCNTs (MRCMW70) in PEG solutions containing BSA for 0 (no dispersion), 5, 10 and 20 min (Table I). Figure 4(a) represents a schematic drawing of the short MWCNTs (MRCMW70) used in the experiments. Figure 4(b) shows representative Raman spectral plots of MWCNT-PEG solutions for different sonication times. All characteristic Raman bands of MWCNTs as well as of PEG were present in all plots. We observed a decrease in Q_CNT as sonication time increased: the average values of Q_CNT were 1.8485 ± 0.0741, 1.5399 ± 0.0257, 1.2685 ± 0.0537, and 1.0676 ± 0.0166, for 0, 5, 10, and 20 min, respectively. This was an anticipated trend because sonication has been shown to induce fracture and fragmentations in MWCNTs.^26,27 Figure 4(c) shows the plot of Q_CNT as a function of sonication time; the data was well-described by an exponential fit (R²= 0.9866). Figure 4(d) shows the absorbance spectra of MWCNT-PEG solutions after 0, 5, 10, and 20 min of sonication. The data shows that the absorbance and sonication time were positively correlated, which implied that longer sonication times led to better dispersion of MWCNTs. This result was expected since sonication for hours or more has been reported effective in dispersing MWCNTs in aqueous solutions.^27,46 Figures 4(e)–4(h) depict optical images of the above described solutions showing MWCNT aggregates (black clusters) becoming smaller as the sonication time increased, which was indicative of an improved dispersion with sonication time.

Additionally, to determine if MWCNTs' geometry played a role in the preservation of MWCNTs quality during sonication, we also evaluated the effect of sonication time on long MWCNTs (MRCMW10) [Fig. 5(a)]. Figure 5(b) shows representative Raman spectral plots of MWCNT-PEG solutions for different sonication times, namely, 0, 5, 10, and 20 min. We again observed consistent decrease in Q_CNT as sonication time increased: the average values of Q_CNT were 1.0187 ± 0.0134, 0.9627 ± 0.0052, 0.9337 ± 0.059, and 0.8924 ± 0.0083, for 0, 5, 10, and 20 min, respectively. Figure 5(c) shows a plot of Q_CNT as a function of sonication time, where we similarly observed an exponential relation between sonication time and Q_CNT (R²= 0.9941). Figure 5(d) shows the absorbance spectra of MWCNT-PEG solutions after 0, 5, 10, and 20 min of sonication. The data consistently showed that longer sonication times led to better dispersion of MWCNTs. Figures 5(e)–5(h) depict optical images of the above described solutions demonstrating MWCNT aggregates (black clusters) becoming smaller as the sonication time increased, again indicative of better MWCNTs dispersion with sonication time.

When comparing between the MWCNTs of two different geometries, Raman spectroscopy showed that for both types of MWCNTs evaluated in this study, Q_CNT decreased exponentially with respect to sonication time. On the other hand, the absorbance spectra at 10 min of sonication and 20 min of sonication were qualitatively similar for either type of MWCNTs. Therefore, 10 min of sonication could render sufficient dispersion while preserving the quality of MWCNTs better as compared to longer sonication times. Ten minutes appeared to be the optimal sonication time for our experiments. Further, from Figs. 4(c) and 5(c), we additionally noted that the rate of change in Q_CNT as a function of sonication time were different. The rate of decrease in Q_CNT was slower for MWCNTs with greater surface area (MRCMW10, Fig. 5) compared to MWCNTs with smaller surface area (MRCMW70, Fig. 4). However, absorbance data showed that both types of MWCNTs had similar absorbance values post-sonication. Therefore, we noted that with similar dispersions, MWCNTs with larger surface area showed better quality preservation during sonication.

We suggest that because MRCMW10 were of larger diameter as compared to MRCMW70, they were under a lower elongation strain,⁴⁷ which rendered them more stable and less likely to be broken by an external energy such as sonication. Larger diameter MRCMW10 are also stiffer than the smaller diameter MRCMW70, which should again make them more stable. Specifically, theory based on a continuum model for a hollow cylinder predicts that bending stiffness, and hence, persistence length of CNTs correlates strongly with tube diameter via the following relationship: bending stiffness = f (nanotube radius).⁴⁸ Further, note that while most other high strength fibers are brittle, CNTs have very large strains at their yield point.⁴⁹ It should also be noted that the dynamic of MWCNTs aggregates in dispersions is complex and likely depends on additional parameters of MWCNTs such specific number of inner tubes, their interspacing as well as external van der Waals tube–tube interactions, which could also play a role in maintaining stable dispersions.²⁶ In our samples (either MRCMW10 or MRCMW70), we did not observe measurable Raman downshift in G bands, which typically is associated with significant strain;^40,47 therefore, we assume that the strain experienced by the CNTs was relatively low.

Probe sonication and bath sonication are the most common sonication methods used at the bench-side as well as in industry. Even though the underlying dispersion mechanism is the same for both methods (refer to Sec. III B), they differ in power and homogeneity and, therefore, in dispersion outcomes. To study the effect of these two sonication conditions on Q_CNT, MWCNTs were dispersed in PEG solutions containing 0.05% BSA for 5 min using both sonication methods (Fig. 6). Figure 6(a) compares the Raman spectra of MWCNT-PEG solutions obtained using probe sonication (top panel) and water-bath sonication (bottom panel). The characteristic Raman bands of both MWCNTs and PEG (D, G, G′, PEG as labeled) were present and the average Q_CNT were 0.9627 ± 0.0052 and 0.9863 ± 0.0038 for probe sonication and water-bath sonication, respectively. This result suggests that water bath sonication was significantly less disruptive to the MWCNTs' structure as compared to probe sonication (p < 0.005); the decrease in Q_CNT was 3.2% for water-bath sonication and 5.5% for probe sonication as compared to untreated MWCNTs.

This result is not surprising since the power input of water-bath sonication is much lower (20–40 W/L) than that of probe sonication (∼20 000 W/L) and higher power input induces more mechanical disruption to the MWCNTs in solution. Figure 6(b) represents the absorbance spectra, where the solid line corresponds to the solution obtained using probe sonication and the dashed line corresponds to the solution obtained using water-bath sonication. We observed that probe sonication gave better dispersion than water-bath sonication as noted by the higher absorption values. Figure 6(c) shows optical images of the above described MWCNTs solutions. Again, we observed that probe sonication (top) yielded better dispersion than water-bath sonication (bottom), noted by the smaller MWCNT aggregates. These results indicated that probe sonication yielded better dispersion than water-bath sonication, and led to a minor 2% decrease in Q_CNT compared to water-bath sonication (p < 0.005). Furthermore, because probe sonication is highly localized and the results are reproducible, as opposed to uneven dispersion typically resulting from water-bath sonication,⁴⁶ we suggest that probe sonication would be a better choice for MWCNTs dispersion in aqueous solutions.

We studied MWCNTs quality preservation post-sonication using Raman spectroscopy. Specifically, we investigated the evolution of Q_CNT as a function of several key parameters, such as MWCNTs geometry, surfactants, sonication time, and sonication method (probe versus water-bath sonication). We demonstrated that Raman spectroscopy was an effective technique in quantifying the quality of MWCNTs in mixed solutions. We determined that as sonication time increased from 0 to 20 min, Q_CNT decreased exponentially, while the dispersion did not change significantly after 10 min of sonication. We also showed that the rate of Q_CNT decrease was different for MWCNTs with different geometry, where MWCNTs with larger surface area exhibited a lower rate of Q_CNT decrease as a function of time. We also observed that surfactants promoted the dispersion of MWCNTs, and preserved the quality of MWCNTs during dispersion, where the protein BSA provided better dispersion as well as protection for MWCNTs as compared to a smaller molecule such as DMSO.

Last, compared to water-bath sonication, probe sonication was shown to be a better sonication method due to its ability to produce better MWCNTs dispersion and only a marginally lower Q_CNT post-sonication. In summary, we described a nondestructive method based on Raman spectroscopy that could be used to determine MWCNTs quality as a function of various treatments, while providing a guide for obtaining CNT-composite aqueous solutions with optimal MWCNTs quality and dispersion for future targeted applications. Last, as discussed above, a similar method could be applied to study the effects of sonication on SWCNTs quality in mixed solutions.

This work was supported by start-up funds awarded to Kuljanishvili and Zustiak by Saint Louis University. Kuljanishvili acknowledges the support from the NSF MRI Program (Award No. 1338021).