Rational Design of a Chalcogenopyrylium-Based Surface-Enhanced Resonance Raman Scattering-Nanoprobe with Attomolar Sensitivity

High sensitivity and specificity are two desirable features in biomedical imaging. Raman imaging has surfaced as a promising optical modality that offers both. Here, we report the design and synthesis of a group of near infrared absorbing 2-thienyl-substituted chalcogenopyrylium dyes tailored to have high affinity for gold. When adsorbed onto gold nanoparticles, these dyes produce biocompatible SERRS-nanoprobes with attomolar limits of detection amenable to ultrasensitive in vivo multiplexed tumor and disease marker detection.


Introduction
Surface-enhanced Raman scattering (SERS) is rapidly gaining interest in the field of biomedical imaging. 1,2,3 By adsorbing a molecule on a noble metal surface, the weak Raman scattering of a molecule (only 1 in ~10 7 photons induces Raman scattering) is massively amplified (enhancement factor 10 7 -10 10 ). 4,5,6 This phenomenon creates a spectroscopic technique that not only has high sensitivity (10 −9 M -10 −12 M limits of detectability), but also the potential for multiplexing capabilities due to the unique vibrational structure of adsorbed molecules. 7,8,9 These characteristics have prompted the use of SERS in a wide array of biomedical imaging applications. 2,10,11,12,13,14,15,16,17 Orders-of-magnitude higher sensitivities (10 −12 -10 −14 M) can be achieved utilizing Raman reporters that are in resonance with the incident laser, thereby producing surface-enhanced resonance Raman scattering (SERRS) nanoprobes. 18,19,20 Absorption of light by biological tissue is minimal in the near-infrared (NIR) window and, as a consequence, most optical biomedical applications use NIR detection lasers. While a great deal of attention has been given to dye molecules that absorb light in the visible region, less work has been devoted to developing Raman reporters with absorption maxima that are resonant with NIR detection lasers. The most common Raman reporters are members of the cyanine class of dyes. 21 Herein we report thiophene-substituted chalcogenopyrylium (CP) dyes as a new class of ultra bright, NIR-absorbing Raman reporters. One notable feature of the pyrylium dyes is the ease in which a broad range of absorptivities can be accessed, and consequently be matched with the NIR light source by careful tuning of the dye's optical properties. Specifically, the large differences in absorption maxima introduced by switching the chalcogen atom is a useful property of this dye class. 22 Another important consideration is the affinity of the reporter for the surface of gold. Since the SERS effect decreases exponentially as a function of distance from the nanoparticle, 23 it is important that the Raman reporter is near the gold surface. The 2-thienyl substituent provides a novel attachment point to gold for Raman reporters. The 2-thienyl group is not only part of the dye chromophore, but also can be rigorously coplanar with the rest of the chromophore. 24 This allows the dye molecules to be in close proximity to the nanoparticle surface, creating a brighter SERRS-signal.

Chalcogenopyrylium dye synthesis and characterization
Cationic chalcogenopyrylium dyes 1-3, with absorption maxima near the 785-nm emission of the detection laser were synthesized as outlined in Figure 1A. The addition of MeMgBr to the known chalcogenopyranones 25 (4,6), followed by dehydration with the appropriate acid (HZ), yields 4-methyl pyrylium compounds (5,7) with the desired counterion (PF 6 − or ClO 4 − ) 22,26,27 The condensation of 7 with N,N-dimethylthioformamide in Ac 2 O, and subsequent hydrolysis of the intermediate iminium salt yields the (chalcogenopyranylidene)acetaldehyde 8, the penultimate compound leading to trimethine chalcogenopyrylium dyes. 22 Condensation of 4-methylpyrylium salt 5 and the (chalcogenopyranylidene)acetaldehyde 8 bearing the desired R groups and chalcogen atom in hot Ac 2 O 26 forms the final dye compounds 1-3 that are substituted with 2-phenyl or 2thienyl groups, and different combinations of chalcogen atoms (S or Se) ( Table 1). The Cl − and Br − counterions of dye 1a were accessed by treating the PF 6 − salts with an Amberlite® ion exchange resin.

SERRS-nanoprobe synthesis and characterization
Chalcogenopyrylium dyes 1-3 were dissolved in dry N,N-dimethylformamide (DMF), at a concentration between 1.0 and 10 mM, and were subsequently used to generate the SERRSnanoprobes. The SERRS-nanoprobes consist of a gold core onto which the SERRS-reporter is adsorbed, which is then protected by an encapsulating silica layer ( Figure 1B, Table 1). The pyrylium based SERRS-nanoprobes were synthesized by encapsulating 60 nm spherical citrate-capped gold nanoparticles via a modified Stöber procedure 28,29,30 in the presence of the reporter. After 25 minutes, the reaction was quenched by the addition of ethanol and the SERRS-nanoprobes were collected through centrifugation. Typically, the as-synthesized SERRS-nanoprobes had a mean diameter of ~100 nm (non-aggregated; Supplementary  Figure 1 and 2).

Effect of counterion on colloidal stability and SERRS-signal
In previous reports, the dye counterion was shown to affect the structural and electronic properties of polymethine dyes 31 and the solubility of chalcogenopyrylium dyes. 32 Since SERRS is highly dependent on these factors, we evaluated the effect of the counterion (Z − ) on the SERRS spectrum, intensity, and colloidal stability of the pyrylium-based SERRSnanoprobes. We compared chloride (Cl − ), bromide (Br − ), perchlorate (ClO 4 − ), and hexafluorophosphate (PF 6 − ) as counterions for chalcogenopyrylium dye 1a. The SERRSnanoprobes were synthesized in the presence of equimolar amounts (10 μM) of CP dye 1a Z − (where Z − = Cl − , Br − , ClO 4 − , or PF 6 − ). The counterion introduces almost no difference in optical properties (e.g. absorption maxima, extinction coefficient). Furthermore, with the exception of the chloride counter-ion, the Raman shifts and intensity of 1a were minimally affected by the different counterions ( Figure 2B). The colloidal stability, however, was shown to be highly counterion dependent ( Figure 2B, Supplementary Figure 1 and Supplementary Table 1). The least chaotropic counterion, Cl − , strongly destabilized the gold colloids and caused aggregation for SERRS-nanoprobes utilizing 1a as a reporter as evidenced by the strong absorption between 700-900 nm. The strongest chaotropic anion, PF 6 − , did not affect colloidal stability during the synthesis of SERRS-nanoprobes as evidenced by the strong absorption at 540 nm and low absorbance between 700 -900 nm (monomeric 60 nm spherical gold nanoparticles have an absorption maximum around 540 nm). Since the PF 6 − anion induced the least nanoparticle aggregation, it was used for further SERRS experiments.

Effect of increased affinity on colloidal stability and SERRS-signal
We also examined the SERRS-signal intensity as a function of the number of sulfur atoms in the dye. Sulfur-containing functionality has been used frequently to adhere molecules to gold, 33 with several reports using thiol or lipoic acid functional groups to add sulfurcontaining functionality. 21,34 In our structures, 2-thienyl groups attached to the 2-and 6positions of the dye were used to bind the dyes to the gold surface. We also explored the impact of the chalcogen atoms in the chalcogenopyrylium core, switching a Se (1a and 2a) to S (1b and 2b). The chalcogen switch was used to increase semi-covalent interactions with the gold surface, and also to create a chromophore that had a more resonant absorption with the 785-nm detection laser (Table 1). Chalcogenopyrylium dyes 1-3 were used at a final concentration of 1.0 μM, which prevented nanoparticle aggregation for dye 3. Figure 3A shows the molecular structures of the chalcogenopyrylium dyes. The SERRS intensity of the different as-synthesized pyrylium-based SERRS-nanoprobes, which were synthesized at equimolar reporter concentrations, were measured at equimolar SERRS-nanoprobe concentrations at low laser power to prevent CCD-saturation (50 μW, 1.0 s acquisition time, 5× objective). We specifically focused on the 1600 cm −1 peak, which corresponds to aromatic ring stretching modes; and is a mode shared by chalcogenopyrylium dyes 1-3. The SERRS-signal intensity of the 1600 cm −1 peak increased significantly as the number of 2thienyl substituents increased ( Figure Table 1). Thus, 3 produced the highest SERRS-signal, which was significantly more intense than 2a/2b or 1a/1b (P<0.05) and 2a/2b were significantly more intense than 1a/1b (P<0.05).
There was a less noticeable, but significant, increase from the chalcogen switch in the core (1a/1b and 2a/2b being significantly different (P<0.05)). This strongly supports the hypothesis that 2-thienyl groups are an effective means of adhering dyes to gold, resulting in brighter SERRS-nanoprobes.

Comparison of CP-dye 3 with a cyanine-based SERRS-reporter
In order to assess the quality of our optimized nanoprobe, thiopyrylium dye 3 and commercially available IR792 ( Figure 4A), which has been previously used to generate surface-enhanced resonance Raman scattering nanoprobes, 35 were studied. A direct comparison of the nanoprobes synthesized in the presence of equimolar (1.0 μM) amounts of 3 and IR792 shows a 5-6-fold higher signal for nanoprobes generated with dye 3 ( Figure  4B). It is interesting to note that a fluorescence background is minimal in the SERRS spectra of the CP-and cyanine-based SERRS-nanoprobes (Supplementary Figure 3). Whereas fluorescence interference would not be expected from chalcogenopyrylium dyes containing heavy chalcogens that enhance intersystem crossing, 36 fluorescence interference could be expected for the cyanine dye IR792. In fact, when equimolar amounts of the CP dyes 1-3 and IR792 were incorporated in silica (without gold nanoparticle), IR792 demonstrated strong fluorescence when excited at 785 nm (50 μW, 1.0 s acquisition time), while minimal fluorescence was observed for CP 1-3. As shown in Figure 4B and Supplementary Figure 3, the fluorescence interference of the cyanine dye IR792 is minimal in its SERRS spectrum. This is due to quenching effects near the surface of the nanoparticle. 37 A concentration series of the as-synthesized SERRS-nanoprobes was generated in triplicate fashion (Supplementary Figure 4) to determine the limit of detection (LOD) of both nanoprobes. Figure 4C shows the LOD for IR792 based nanoprobes to be 1.0 fM, while 3based nanoprobes had a 10-fold lower LOD, 100 aM. To our knowledge this is the lowest reported LOD utilizing a biologically relevant NIR excitation source. We also evaluated the serum stability of the 3-based SERRS-nanoprobe. The SERRS-nanoprobe was shown to be serum stable (e.g. no significant difference between t = 1 h and t = 48 h) for at least 48 hours (Supplementary Figure 5; Supplementary Methods). This is supported by a study by Thakor et al. who have shown that SERS-nanoparticles of similar size and composition remain stable in vivo for more than 2 weeks. 38

In vivo comparison of EGFR-targeted CP3-or IR792-SERRS-nanoprobes
The ability of our SERRS-nanoprobe to delineate tumor tissue in vivo was assessed by utilizing CP dye 3 and IR792-based SERRS-nanoprobes functionalized with an epidermal growth factor receptor (EGFR)-targeting antibody. Equimolar amounts (15 fmol/g) of these two EGFR-targeted nanoprobes were injected intravenously into athymic nude mice which had been inoculated two weeks prior with the EGFR-overexpressing cell line A431 (1 × 10 6 cells). After 18 hours, the skin around the tumor was carefully peeled back and multiplexed Raman imaging the tumor site and surrounding tissue was performed ( Figure 5-6). A Raman map was generated and the signals from the multiplexed SERRS-nanoprobes were deconvoluted by applying a direct classical least square algorithm (DCLS). 7 The SERRSsignal from both nanoprobes was more intense for the tumor site than for the surrounding tissue, showing that the EGFR-targeted SERRS-nanoprobes had selectively localized at the tumor site. The SERRS-signal intensity at the tumor mass revealed a 3× higher signal density for the 3-based SERRS-nanoprobes than for the otherwise identical IR792-based SERRS-nanoprobes. Ex vivo multiplexed Raman imaging of the tumor showed Raman signal of the EGFR-targeted SERRS-nanoprobes throughout the tumor with the exception of a hypointense Raman region in the center of the tumor. H&E and immunohistochemical staining for EGFR was performed (Supplementary Methods) and revealed that the hypointense Raman region corresponded with an area of necrosis, which explains the lack of SERRS-nanoprobe accumulation and decreased Raman signal. In addition, to validate EGFR targeting, we injected A431-tumor bearing mice with cetuximab (50 pmol/g) 3 hours prior to injection with the EGFR-targeted SERRS-nanoprobes. Pre-blocking of EGFR by cetuximab resulted in decreased accumulation of the EGFR-targeted SERRS-nanoprobes within the tumors of animals that were injected with cetuximab prior to EGFR-targeted SERRS-nanoprobe injection as compared to animals that were injected with EGFR-targeted SERRS-nanoprobes and were not pre-injected with cetuximab (Supplementary Figure 6).

Discussion
Effective biomedical imaging requires low limits of detection and high specificity for biological targets. Raman imaging has surfaced as an optical imaging modality that has the promise to enable both. While the Raman effect is relatively weak (1 in 10 7 photons), 3-5 the Raman scattering cross section of a molecule can be massively amplified by noble metal surfaces. Here, we demonstrated that rational SERRS-reporter design afforded SERRSnanoprobes with unprecedented limits of detection: 100 attomolar. This is to the best of our knowledge the lowest reported limit of detection at near-real-time detection (≤2.0 s acquisition times) for SERRS-nanoprobes that are compatible with a NIR light source. As a comparison non-resonant SERS-nanoprobes are in the 0.1-1.0 pM range (1,000-10,000-fold less sensitive) 2 , while reported detection limits of SERRS-nanoprobes are >17 fM at near real-time detection. 39 Others have reported a 0.4 fM detection limit, however, this was acquired through cumulative data acquisition with an acquisition time ≥60s, which is not practical for biomedical imaging applications. 35 We believe the unprecedented limit of detection of our novel SERRS-nanoprobe is due to several factors. First, we demonstrate that rational design and optimization of the SERRSreporter is important to achieve efficient "loading" on the nanoparticle. Our results demonstrate that the counterion and gold surface affinity are important considerations. For instance, while the chaotropic PF 6 − anions stabilized the dye-nanoparticle system during silica shell formation in ethanol, the system becomes more destabilized with Cl − (more kosmotropic) ions present. Chloride-induced aggregation of colloidal dispersions in relation to SERS has been studied. Natan et al. demonstrated that the strongest enhancements were obtained from aggregates with effective diameters of less than 200nm and aggregates with sizes >200nm did not generate appreciable SERS intensities. 40 The aggregates that were induced by the chloride counterion in our system were >200 nm (Supplementary Figure 2), which might explain the reduced SERRS-signal when chloride is used as a counterion.
Others have shown that the kosmotropic chloride-ion could induce reorientation of the dye on the surface, which could also contribute to the reduced SERRS intensities. 41 However, while we did observe a decrease in the SERRS-signal intensity when chloride is present, we did not find any appreciable differences between the Raman spectra of the dyes when different counterions were use, which would have been expected if the molecule had reoriented on the surface. Since the most chaotropic counterion, PF 6 − , induced the least aggregation and generated robust SERRS-signal intensities, we used PF 6 − as a counterion.
Next, we showed that an increase in affinity of the SERRS-reporter for the gold nanoparticle surface via incorporation of 2-thienyl functional groups considerably increased the SERRSsignal without inducing aggregation. Others have reported the functionalization of NIR dyes with thiol or lipoic acid functional groups. In contrast to a 2-thienyl substituent, thiol and lipoic acid functional groups offer no benefit to the optical properties of the dye, and as a tether, do not allow the dye to be as close to the gold surface. Moreover, based on the absorption spectra of reported lipoic-acid modified cyanine dye-gold nanoparticle conjugates, it is clear that lipoic-acid modified dyes promote aggregation. 21,34 Finally, the strategy chosen to stabilize the SERRS-nanoprobe is a key factor. Others have reported using either surfactants or thiolated-polymers to stabilize their SERRSnanoparticles. 35,39,42 However, such stabilizing agents compete with the SERRS-reporter for the surface of the nanoparticle, which leads to relatively low SE(R)RS-signal. We achieved very low limits of detection by using a primerless silication procedure in which the silica not only served as a stabilizing agent, but also as a matrix to contain our optimized CP-based SERRS-reporter. Since silica has much lower affinity for the gold than the applied SERRS-reporters, attomolar limits of detection were achieved.
The chalcogenopyrylium dyes represent a new class of SERRS-reporters. Selection of the right combination of chaotropic counterions and increased affinity of the SERRS-reporter for the gold nanoparticle's surface produces stable SERRS-nanoprobes with exceptionally low limits of detection (attomolar range). The low limit of detection (i.e. close to single nanoparticle detection) in combination with the high resolution of Raman imaging, enables highly sensitive and specific, near-real-time tumor delineation and, as a result of the fingerprint like spectra of the different SERRS-nanoprobes, can offer multiplexed disease marker detection in vivo.

Chalcogenopyrylium dye synthesis and characterization
All reactions were done open to air unless otherwise noted. Concentration in vacuo was performed on a rotary evaporator. NMR spectra were recorded at 300 or 500 MHz for 1 H and at 75.5 MHz for 13 C with residual solvent signal as internal standard. If a mixture of CD 2 Cl 2 and CD 3 OD was used to acquire 1 H NMR, the peak for CH 2 Cl 2 was used as the internal standard. UV/VIS-near-IR spectra were recorded in quartz cuvettes with a 1-cm path length. Melting points were determined with a capillary melting point apparatus and are uncorrected. Non-hygroscopic compounds have a purity of ≥ 95% as determined by elemental analyses for C, H, and N. Experimental values of C, H, and N are within 0.3% of theoretical values. 13 C NMR was not recorded for pyrylium dyes due to limited solubility in common NMR solvents. Chalcogenopyranones 25 4 and 6 were made according to literature procedures as were 4-methylchalcogenopyrylium and 4-methylchalcogenobenzopyrylium compounds 5 and 7. 22,25,26,27   Cl − : The hexafluorophosphate salt (50 mg) was converted to the chloride salt by treating with Amberlite® IRA-400 chloride form (200 mg) in a 1:1 CH 2 Cl 2 :MeOH mixture (3.0 mL). This process was repeated two more times after which the product was dissolved in CH 2 Cl 2 , washed with water, the organic layer dried with Na 2 SO 4 , filtered over Celite®, and Br − : The hexafluorophosphate salt (50 mg) was converted to the bromide salt by treating with Amberlite® IRA-400 bromide form (200 mg) in a 1:1 CH 2 Cl 2 :MeOH mixture (3.0 mL). This process was repeated two more times after which the product was dissolved in CH 2 Cl 2 , washed with water, the organic layer dried with Na 2 SO 4 , filtered over Celite®, and after concentration recrystallized from CH 3

SERRS-nanoprobe synthesis
Gold nanoparticles were synthesized through addition of 7.5 mL 1% (w/v) sodium citrate to

SERRS-nanoprobe characterization
The as-synthesized SERRS-nanoprobes were characterized by transmission electron microscopy (TEM; JEOL 1200ex-II, 80 kV, 150,000× magnification) to study the SERRSnanoprobe structural morphology. The size and concentration of the SERRS-nanoprobes were determined on a Nanoparticle Tracking Analyzer (NTA; Malvern Instruments, Malvern, UK). Absorption spectra to determine possible nanoparticle aggregation (typically detectable at wavelengths > 600 nm) were measured on an M1000Pro spectrophotometer (Tecan Systems Inc. San Jose, CA). Finally Raman spectra were acquired on a Renishaw InVIA system equipped with a 785-nm laser (Renishaw Inc, Hoffman Estates, IL). All measurements were performed at a laser power of 50 μW (1.0 s acquisition time, 5× objective).

SERRS-nanoprobe limit of detection
SERRS-nanoprobes were synthesized as described above in the presence of an equimolar (1.0 μM) amount of 3 or IR792. SERRS imaging to determine the limit of detection was performed at 100 mW (2.0 s acquisition time (StreamLime™), 5× objective) on a phantom that consisted of a serial diluted IR792-or chalcogenopyrylium dye 3-based SERRSnanoprobe redispersed in 10 μL water (concentration range 3000 -0.003 fM; n=3). The Raman maps were generated by WiRE 3.4 software (Renishaw) by applying a direct classical least square (DCLS) algorithm. The Raman image was analyzed with ImageJ software and plotted in GraphPad Prism (GraphPad Software Inc., La Jolla, CA).

Animal studies
All animal experiments were approved by the Institutional Animal Care and Use Committees of Memorial Sloan Kettering Cancer Center.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.

Figure 5. Comparison between EGFR-targeted IR792-or 3-based SERRS-nanoprobes in an A431 tumor xenograft
Female nude mice (n=5) bearing A431 xenograft tumors were injected intravenously via tail vein with an equimolar amount of EGFR-antibody (cetuximab)-conjugated IR792-and CP 3-based SERRS-nanoprobes (15 fmol/g per probe; total injected dose: 30 fmol/g). After 18 hours, the tumors were imaged in situ by Raman (10 mW, 1.5 s acquisition time, 5× objective). The chalcogenopyrylium dye 3-based SERRS-nanoprobe (red) provided ~3× more contrast than the IR792-based SERRS-nanoprobe (cyan) (22.442 cps/cm 2 versus 7.313 cps/cm 2 , respectively). All scale bars represent 2.0 mm. The excised tumor was scanned by Raman imaging (10 mW, 1.5 s acquisition time, 5× objective) and subsequently fixed in 4% paraformaldehyde and processed for H&E staining and anti-EGFR immunohistochemistry. With the exception of a hypointense Raman region in the center of the tumor, the tumor homogenously expressed EGFR and the EGFR-targeted SERRS-nanoprobes had accumulated throughout the tumor. The hypointense Raman area corresponds to a highly necrotic region within the tumor, which explains the lack of SERRSnanoprobe accumulation and decreased Raman signal. All scale bars represent 1.0 mm.