Optimizing Quantum Dot Probe Size for Single-Receptor Imaging

Abstract

Quantum dots (QDs) are nanocrystals with bright fluorescence and long-term photostability, attributes particularly beneficial for single-molecule imaging and molecular counting in the life sciences. The size of a QD nanocrystal determines its physicochemical and photophysical properties, both of which dictate the success of imaging applications. Larger nanocrystals typically have better optical properties, with higher brightness, red-shifted emission, reduced blinking, and greater stability. However, larger nanocrystals introduce molecular-labeling biases due to steric hindrance and nonspecific binding. Here, we systematically analyze the impact of nanocrystal size on receptor labeling in live and fixed cells. We designed three (core)shell QDs with red emission (600–700 nm) and crystalline sizes of 3.2, 5.5, and 8.3 nm. After coating with the same multidentate polymer, hydrodynamic sizes were 9.2 nm (QD_9.2), 13.3 nm (QD_13.3), and 17.4 nm (QD_17.4), respectively. The QDs were conjugated to streptavidin and applied as probes for biotinylated neurotransmitter receptors. QD_9.2 exhibited the highest labeling specificity for receptors in the narrow synaptic cleft (~20–30 nm) in living neurons. However, for dense receptor labeling for molecular counting in live and fixed HeLa cells, QD_13.3 yielded the highest counts. Nonspecific binding rose sharply for hydrodynamic sizes larger than 13.3 nm, with QD_17.4 exhibiting particularly diminished specificity. Our comparisons further highlight needs to continue engineering the smallest QDs to increase single-molecule intensity, suppress blinking frequency, and inhibit nonspecific labeling in fixed and permeabilized cells. These results lay a foundation for designing QD probes with further reduced sizes to achieve unbiased labeling for quantitative and single-molecule imaging.

Graphical Abstract

Quantum dots (QDs) are semiconductor nanocrystals with bright photoluminescence, narrow and tunable emission bands, and high photostability.^1-3 After coating with hydrophilic organic ligands, QDs can be homogeneously dispersed in biological media and efficiently attached to bioaffinity molecules such as antibodies and nucleic acids. This combination of attributes makes QDs particularly useful as optical probes for studying molecular dynamics in living cells through single-molecule tracking, which requires thousands of cycles of excitation and emission.^3-6 QDs have also enabled absolute molecular counting of proteins and mRNA in single cells and tissues, which has been challenging using dimmer and less photostable organic fluorophores.^7,8 These two fields of dynamic single-molecule tracking and static single-molecule counting apply the same QD probes in different experimental and analytical workflows, each of which are currently enhancing molecular and cellular analyses in the life sciences,⁹ with potential benefits in clinical in vitro diagnostic assays.¹⁰

A primary design parameter for QDs is the size of the nanocrystal, which is typically between 2 and 10 nm. Most QD crystalline domains are composed of epitaxial (core)shell structures such as (CdSe)Cd_xZn_1-xS, for which the core size and shell thickness are independently tunable and contribute collectively to the total nanocrystal size. Increasing either the core size or shell thickness will enhance the optical and electronic properties of these quantum-confined materials. This is because, for a fixed composition, larger cores have narrower electronic bandgaps with emission and absorption at longer wavelengths at which autofluorescence from cells is reduced. Increasing core size also increases brightness,¹¹ as extinction coefficients (ε) are approximately proportional to nanocrystal core volume.¹² The function of the shell is to boost quantum yield (QY) and stability due to electronic insulation that occurs when the bandgap material of the shell is wider than that of the core.^13,14 Increasing shell thickness leads to progressive improvements in signal intensity¹⁵ and also improves temporal emission uniformity at the single-molecule level, with fewer “off” dark events through the process of “blinking”.^16,17 For these reasons, larger nanocrystal sizes overall are preferred for single-molecule tracking and counting applications, which require bright, stable, and nonblinking probes to maximize the signal-to-noise ratio for molecular detection and localization at the fastest rate.¹⁸ However, together with contributions from the organic coating, larger nanocrystal sizes also exhibit larger hydrodynamic dimensions, which increase steric hindrance when the QD functions as a molecular label. As a result, larger QDs may prevent target labeling in molecularly crowded microenvironments such as a neuronal synapse^19,20 and the cytoplasm,⁸ possibly leading to off-target “nonspecific” labeling. Therefore, smaller probes are preferred physically to maximize labeling specificity with minimal bias. Efforts to minimize the QD hydrodynamic size to date have primarily focused on shrinking the size of the organic surface coatings based on ligands and polymers.^21,22 However, size trade-offs for the nanocrystalline domain have yet to be balanced and refined.

Here, we evaluate the impact of nanocrystal size on the performance of QDs for high-resolution single-receptor imaging in live and fixed cells. We designed, synthesized, and characterized (core)shell (Hg_xCd_1−xSe)Cd_yZn_1−yS QDs with diameters of 3.2, 5.5, and 8.3 nm (Figure 1a). Traditional wavelength tuning through nanocrystal size alone results in large differences in emission wavelength, complicating performance comparisons when both cellular autofluorescence intensity and camera quantum efficiency are dependent on wavelength, particularly obscuring smaller QDs which emit in the green spectrum.⁷ Therefore, for the smallest QDs, we used wider bandgap cores composed of ternary alloy Hg_xCd_1−xSe so that all QDs emit in the red spectrum (600–700 nm) with high QY. We coated each QD identically with the multidentate polymer ligand polyacrylamido(histamine-co-triethylene glycol-co-azidotriethylene glycol) (P-IM-N₃; Figure 1b,c) which appends short hydrophilic oligoethylene glycol (OEG) groups to the surface to minimize total hydrodynamic size and maintain a homogeneous size distribution.²³ The resulting QDs have hydrodynamic diameters of 9.2, 13.3, and 17.4 nm, and are referred to as, respectively, QD_9.2, QD_13.3, and QD_17.4. The P-IM-N₃ polymer also facilitates efficient bioconjugation via azide–alkyne click chemistry, allowing facile conjugation to streptavidin (SAv) functionalized with dibenzylcyclooctyne (DBCO) to generate probes for recombinant biotinylated receptors. This receptor labeling scheme enables a short linker length compared with receptor-targeted antibodies, with a fast association rate and nearly covalent bond strength. We applied QD–SAv conjugates to densely label and image plasma membrane receptors engineered with extracellular biotin tags expressed in HeLa cells (Figure 1d). We counted labeling events both in live cells and in fixed, permeabilized cells, quantified specificity and labeling density, and interpreted outcomes based on the photophysical and photochemical properties of the three nanocrystals. We further applied these QDs to sparsely label and image receptors on live hippocampal neurons in which biotinylated receptors were endogenously localized within synapses, which are intercellular junctions with ~20–30 nm spacings (Figure 1d). We measured the fraction of QDs that access synaptic receptors and interpreted outcomes based on nanocrystal size. In all applications, we targeted a glutamate receptor subunit (GluR2) of the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR),²⁴ which we ectopically expressed in HeLa cells and orthotopically expressed in neurons. Our results show that smaller nanocrystal sizes are optimal for live-cell receptor labeling and tracking within neuronal synapses, which is consistent with previous reports.^19,20 However, for applications requiring dense molecular counting, the smallest QD_9.2 needs further engineering and development to eliminate biases deriving from low intensity and intermittent emission and to minimize nonspecific interactions with intracellular nuclear domains of fixed cells.

Figure 1.

Quantum dot molecular probe design for labeling and imaging of membrane receptors. (a) Schematic depiction of (core)shell QD syntheses to prepare three fluorescent QDs with different sizes, all emitting at red wavelengths. (b) Schematic depiction of three QDs in (a) after coating with P-IM-N₃ polymer. (c) Schematic depiction of QD coating with P-IM-N₃ polymer and conjugation to DBCO-functionalized SAv. (d) Schematic depiction of different molecular labeling scenarios evaluated in this manuscript using QDs with different sizes as fluorescent probes.

RESULTS AND DISCUSSION

Quantum Dot Probe Physical Properties.

We synthesized three quasispherical (core)shell (Hg_xCd_1−xSe)Cd_yZn_1−yS QDs with diameters of 3.20 ± 0.40 nm, 5.48 ± 0.68 nm, and 8.31 ± 0.98 nm (mean ± standard deviation) measured via transmission electron microscopy (TEM) (Figure 2a,b). Each of these materials was designed and independently optimized for high brightness and stability, with a minimum of 2.4 monolayers of insulating shell grown epitaxially. In particular, the 3.2 nm core is our best current design for small QDs, with mercury incorporated into the core to shift its emission to the red spectrum, and with core and shell dimensions carefully chosen to balance absorption, QY, stability, and total size. Each QD was coated with the azide functionalized multidentate polymer P-IM-N₃, resulting in three QDs with mean hydrodynamic diameters of 9.2 nm (QD_9.2), 13.3 nm (QD_13.3), and 17.4 nm (QD_17.4) measured by gel permeation chromatography (GPC) (Figure 2c) calibrated by globular protein size standards. GPC chromatograms show that all three QDs are >95% monomeric in phosphate buffered saline (PBS), and that the polymer coating introduces a radial coating thickness of 3.0, 3.9, and 4.5 nm around the core for QD_9.2, QD_13.3, and QD_17.4, respectively. The differences in thickness likely reflect slightly different polymer conformations on these surfaces due to the differences in surface curvature that span a range of more than 6-fold.²⁵ We also confirmed these QD size distributions using fluorescence correlation spectroscopy (FCS) and measured similar size ranges (Figure 2d). Example FCS curves are shown in Figure S1.

Figure 2.

Characterization of QD dimensions and functionality. (a) Representative TEM images of three QDs used in the study. Scale bar = 20 nm. (b) Size histograms of QDs measured by TEM. Each distribution is fit to a Gaussian function where d indicates diameter at the centroid of each Gaussian with indicated standard deviation. (c) Hydrodynamic size of P-IM-N₃-coated QDs measured by gel permeation chromatography (GPC) with indicated sizes determined from globular protein size standards. O.D. = optical density. A.U. = arbitrary units. (d) Hydrodynamic diameter of QDs in the absence and presence of 1% BSA in PBS measured by fluorescence correlation spectroscopy (FCS). Example FCS curves are shown in Figure S1. X-axis error bars are the s.d. of diameter distributions from TEM. Y-axis error bars are s.d. from FCS measurements (e) Gel electrophoresis analysis of conjugation reactions between QDs and SAv, indicating that QD–SAv conjugates bind to biotinylated DNA (B-DNA) based on band shifts. In each gel, the QD alone (lane 1) migrates a longer distance than QD–SAv (lane 2), reflecting the larger size of QD–SAv. QD–SAv-B-DNA migrates a longer distance than both due to the higher net negative charge of DNA.

To determine if core size impacts nonspecific binding to a representative biological molecule, each P-IM-N₃-coated QD was incubated in bovine serum albumin (BSA, 1% w/v) in PBS, and hydrodynamic size was measured via FCS. Binding to BSA was QD-size-dependent, reflected by a progressive increase in radial expansion of size with increasing QD core size due to adsorption of BSA (Figure 2d). BSA increased the QD diameter by 3.8 nm for QD_9.2, 4.8 nm for QD_13.3, and 10.8 nm for QD_17.4. These dimensions are important because BSA is used at this same concentration as a colloidal blocking agent for applications in live and fixed HeLa cells, described below, so the molecularly targeted QD probes may behave with these expanded dimensions. Because BSA has a hydrodynamic diameter of ~7 nm, with approximate 4 nm × 4 nm × 14 nm dimensions,²⁶ an increase of 3.8 nm for the smallest QDs suggests the adsorption of one or fewer BSA molecules, possibly indicating a “soft” weakly adsorbed corona or indicating that a portion of BSA is buried within the coating polymer.^27-29 The much thicker BSA corona on QD_17.4 potentially indicates a complete monolayer of adsorption. The greater adsorption of proteins to larger particles can be attributed to the larger total surface area which increases adsorption rate, and to their lower surface curvature which allows more points of adsorptive contact.^27-29

In order to compare the performance of the QDs for labeling biotinylated receptors on cells, we conjugated the azide-functional QDs to SAv that was covalently modified with DBCO, resulting in a covalent bond mediated by copper-free click chemistry,^30,31 confirmed by a reduction in electrophoretic mobility deriving from a larger size of the conjugate (Figure 2e). Efficient binding to biotin was verified by mixing each of the QD–SAv conjugates with biotin-terminated single-stranded DNA, which resulted in more rapid electrophoretic migration deriving from the anionic charge on DNA and a disappearance of the QD–SAv band. We verified the specificity of QD–SAv conjugates toward surface-bound biotin using glass surfaces modified with polyethylene glycol (PEG) terminated with biotin with inhibition by free biotin, confirming dependence on biotin (Figure S2).

Quantum Dot Photophysical Properties.

We characterized the photophysical properties of each of the QDs after polymer coating. Shown in Figure 3a, the emission peak maxima were 646, 608, and 685 nm for QD_9.2, QD_13.3, and QD_17.4, respectively, with corresponding QY of 23%, 46%, and 28% in PBS, which are all in a similar range. The unconventional relationship between fluorescence wavelength and size derives from the high content of mercury in QD_9.2,³² used to shift the emission peak to be centered between that of the other two QDs. The emission band of QD_9.2 is notably much broader than that of the other two, which is common for smaller QDs with higher vibronic coupling and for alloys which exhibit some heterogeneous broadening due to a distribution of compositions.^11,33 The range of QYs is consistent with our previous reports of different batches of QDs coated with the same polymer.²³

Figure 3.

Characterization of QD photophysical properties at the ensemble level. (a) Emission spectra of QD_9.2, QD_13.3, and QD_17.4, with corresponding emission band-pass filters used for imaging in microscopy experiments. Band-pass filters cover 88%, 92%, and 90% of the emission spectra of QD_9.2, QD_13.3, and QD_17.4, respectively. A.U. = arbitrary units. (b) Extinction coefficient spectra of the three QDs. (c) Relative ensemble brightness for different excitation wavelengths for the three QDs in PBS.

Unlike their relatively similar emission wavelength ranges and QY, extinction coefficients and brightness are drastically dissimilar across the three QD sizes. Figure 3b shows the large dissimilarity in extinction that derives from size. At 405 nm wavelength excitation, a common excitation wavelength for QDs, ε of QD_13.3 is 3 times larger than that of QD_9.2 and ε of QD_17.4 is 27 times larger than that of QD_9.2. Most live-cell applications use longer excitation wavelengths for which phototoxicity is reduced.^34,35 At 561 nm wavelength excitation used in most of this work, ε for QD_13.3 and QD_9.2 are nearly equivalent; while ε of QD_17.4 is 24 times greater than that of QD_9.2 (Figure 3b and Figure S3). These differences in extinction coefficient are the primary determinates of brightness differences. The spectra of relative brightness, B_rel, are shown in Figure 3c at the ensemble level measured by the excitation spectra of the QDs.^11,36 Here

$$B_{\text{rel}}=\epsilon \times \text{QY}$$ (1)

At 561 nm wavelength excitation, QD_13.3 is 1.8 times brighter than QD_9.2 and QD_17.4 is 29 times brighter than QD_9.2, assuming measurements are performed under identical experimental conditions.

We next characterized the brightness and blinking characteristics of the three aqueous QDs at the single-molecule level using wide-field fluorescence microscopy. Single-QD images in Figure 4a show the three QDs absorbed sparsely to a glass surface, immersed in PBS, and excited with 561 nm wavelength light. The relative intensities were generally similar to those at the ensemble level, with increasing intensities for increasing size. This trend is consistent with our previous report showing that brightness at the ensemble level correlated with QD intensity at the single-molecule level in the “on” blinking state.¹¹ Representative intensity time-traces of individual QDs are shown in Figure 4b and plotted as an integrated intensity histograms in Figure 4c with fitted functions to classify spots as single QDs or clusters based on their intensity distributions.^7,11,18 On average, QD_13.3 was 2.2 times brighter than QD_9.2, and QD_17.4 was 19 times brighter than QD_9.2 for single-QD traces averaged over time (Figure 4d), results similar to those measured at the ensemble level. In addition, as expected, the off-time fraction of blinking decreased with increasing QD size, consistent with the thicker shell for the larger sizes that eliminates electronic trap states (Figure 4e). These results confirm that the larger QDs are brighter and more commonly in the “on” blinking state than the smaller QDs.

Figure 4.

Characterization of QD photophysics at the single-molecule level. (a) Representative fluorescence micrographs of QD_9.2, QD_13.3, and QD_17.4 on glass surfaces. Red arrow indicates a single QD spot. Scale bar = 5 μm. (b) Examples of intensity time traces of the three QDs with different sizes. (c) Gaussian fits of intensity histograms shown in (b), indicating single QD blinking behavior. (d) Mean fluorescence intensity of single QDs. A.U. = arbitrary units. (e) Average off-fraction of three QDs at the single-QD level. In (d) and (e), N = 74, 86, and 74 for QD_9.2, QD_13.3, and QD_17.4, respectively. Bar graphs indicates mean ± s.e.m. * indicates p < 0.05; ** indicates p < 0.005.

Labeling Receptors on Live HeLa Cells.

We first evaluated the impact of QD size on quantitative labeling and counting of a plasma membrane receptor on live HeLa cells, which are immortalized human epithelial cells. We chose to target the GluR2 subunit of the AMPA receptor^19,20 so that the same construct could be used to study its orthotopic expression in neurons in later experiments described below. To directly quantify specific and nonspecific binding, we developed a cell expression system comprising a mixture of cells expressing and not expressing the target of interest. HeLa cells, which natively do not express biotinylated GluR2 (B-GluR2), were transiently transfected with a single plasmid containing an AviTag peptide linked to GluR2, enhanced green fluorescent protein (eGFP) as a transfection reporter, and biotin ligase (BirA). BirA conjugates biotin to the lysine on the 15-amino acid AviTag peptide (Figure 5a). QD–SAv conjugates were then added to the cells to label B-GluR2, and single QD spots were counted from wide-field fluorescence microscopy images collected through highly inclined and laminated optical sheet (HILO) imaging that illuminates only a thin plane. Specific and nonspecific binding were determined using the same images, exploiting the mixed population of transfected and nontransfected cells identified based on GFP expression (Figure 5b-d).

Figure 5.

Evaluation of GluR2 membrane receptor labeling density and specificity using three QD sizes in live HeLa cells. (a) Schematic of plasmid used to induce expression of biotinylated GluR2 (B-GluR2) in HeLa cells. (b) Schematic of cell transfection and QD–SAv labeling workflow showing B-GluR2-expressing cells in green reflecting GFP reporter expression. (c) Schematic of nonspecific binding of QD–SAv to HeLa cells not expressing B-GluR2. (d) Schematic of specific binding of QD–SAv to B-GluR2 on transfected HeLa cells. (e) Representative images of QDs binding to GFP-positive and GFP-negative cells after 2 h treatment with 10 nM QD–SAv at 37 °C. All images show the same image acquisition conditions except for the red-outlined inset for QD_17.4, for which intensity is reduced to observe individual QDs. Nuclei are shown in blue; GFP is shown in green; scale bar = 10 μm. (f) Number of QD–SAv bound per cell for GFP positive and negative cells. Box plots show 25/75 percentile; red lines indicate mean values; whiskers indicate s.d.; * indicates p < 0.05; ** indicates p < 0.005; N.S. indicates p > 0.05. (g) Ratio of number of QDs bound to GFP-positive cells divided by the number of QDs bound to GFP-negative cells as shown in (f). Bar graphs indicates mean ± s.e.m. (h,i) QD number detected per cell based on spot detection thresholds for (h) GFP-positive cells and (i) GFP-negative cells in (f). (j) Ratio of number of QDs bound to GFP-positive cells divide by number of QDs bound to GFP-negative cells based on spot detection thresholds as shown in (h) and (i). Plots show mean ± s.e.m. at each threshold value. (k) Time-course of fluorescence intensity of QDs at pH = 4 and 7.2 measured in buffered solutions. (l) Relative fluorescence intensity of QDs at indicated pH after 2 h incubation in buffered solution. In (f) and (g), for GFP-negative cells, N = 21, 14, and 24, respectively, for QD_9.2, QD_13.3, and QD_17.4. For GFP-positive cells, N = 51, 26, and 16, respectively, for QD_9.2, QD_13.3, and QD_17.4. In (k) and (l), points indicate mean ± s.d., N = 3. Fluorescence intensity was normalized to fluorescence intensity at pH 7.2. Note that for all three QDs, we arbitrarily selected cells with similar B-GluR2 expression level based on GFP intensity (Figure S5). Hence, the differences in nonspecific and specific binding between the QDs are solely due to differences in QD size and not due to target protein expression levels. In (h), and (i), plots show mean ± s.d. at each threshold value.

We used 10 nM QD–SAv conjugates, a typical concentration for comprehensive labeling of protein targets for both immunofluorescence imaging and flow cytometry.^7,37,38 A binding time course for QD_9.2–SAv in a single z-plane showed that B-GluR2-specific labeling events in GFP-positive cells increased with incubation time from 0.8, to 3.1, to 3.7 QD per cell for 30, 60, and 120 min, respectively (Figure S4). In comparison, fewer than 1 nonspecific QD per cell was observed for GFP-negative cells. We chose the 120 min incubation time point to evaluate binding of all QD sizes and found that both of the small QD_9.2–SAv and QD_13.3–SAv exhibited a similarly low level of nonspecific binding with little difference between the two (p > 0.05) (Figure 5e,f). However, for specific binding to GFP-positive cells, there were 2.5 times more QD_13.3–SAv than QD_9.2–SAv. As a result, QD_13.3–SAv yielded the highest signal-to-background ratio of the QDs, defined as the number of QDs per biotinylated GluR2-expressing cells divided by that of nonexpressing cells (Figure 5g). The number of QD_13.3–SAv bound was similar to that for Alexa594–SAv (Figure S6) in GFP-positive cells, demonstrating that the QD_13.3 label does not impact the ability of SAv to bind to biotinylated GluR2 on live HeLa cells. This is consistent with our recent work⁷ showing that QDs with similar size as QD_13.3 can label epidermal growth factor receptor (EGFR) with labeling density proportional to the number of membrane receptors.⁷

We hypothesized that the smaller number of counts of QD_9.2 compared to QD_13.3 could be due to several reasons, including their differences in brightness (Figure 4d), blinking (Figure 4e), and stability toward chemical microenvironments. To analyze the impact of single QD brightness, we varied the detection threshold of the spot detecting algorithm.³⁹ Shown in Figure 5h, the lower brightness of QD_9.2 causes the spot detection sensitivity to exhibit a greater dependence on threshold than QD_13.3, reflected by a greater magnitude of slope. Therefore, arbitrary thresholding can introduce spot detection biases that can result in more substantial undercounting of QD_9.2 relative to QD_13.3. We used a stringent threshold of 28 to prevent autofluorescence events from registering as counts, so thresholding biases do account for some of the fraction of undercounting by the observed factor of 2.5. We also calculated QD counts on GFP-negative cells with different detection thresholds (Figure 5i) in order to calculate the signal-to-background ratio across different detection thresholds (Figure 5j). The result was that across all thresholds, the signal-to-background ratio of QD_13.3 was better than that of QD_9.2 and QD_17.4, with increasing disparity with increasing threshold. We emphasize that even with a judiciously chosen threshold, specific counts here (tens of QD or Alexa594) are not equivalent to the absolute expression level of the biotinylated receptor, as HILO imaging illuminates only a thin plane through the cells and many of the molecular targets are inaccessible to the probes because they are intracellular in this ectopic expression system (see below). The counts are consistent with previous studies in which comprehensive 3D counting was performed across all image planes of individual cells labeled with receptor-targeted QD-ligand conjugates.⁷ In these previous studies, a single image plane yielding ~40 detected spots equated to ~1000 comprehensively counted spots in 3D across the entire cell.

We further evaluated the impact of chemical microenvironment. During the 120 min incubation, QD–SAv bound to GluR2 can be internalized into endosomes, where the pH (~6) is significantly reduced compared to the extracellular medium (pH ~ 7.4). This pH is expected to reduce further as endosomes mature to late endosomes (pH ~ 5–6), and fuse with lysosomes (pH ~ 4.7).⁴⁰ We determined that in the absence of cells acidic pH diminishes the brightness of all of the QDs in buffered solutions at a magnitude dependent on size (Figure 5k,l and Figure S7). QD_9.2 exhibited the most rapid reduction in brightness with decreasing pH, which is consistent with its thinnest shell and lowest degree of electronic insulation. However, the reduction in brightness at reduced pH was statistically similar for QD_9.2 and QD_13.3 at the 120 min time point that was used for cell imaging (p > 0.5, Figure 5j). Therefore, the lower number of specific counts for QD_9.2 compared to QD_13.3 (Figure 5f) likely arose from intrinsic differences in brightness (Figure 4d) as well as blinking that causes a larger fraction of QD_9.2 to be in the “off” blinking state during the period of image acquisition (Figure 4e). Notably, QD_17.4–SAv exhibited the strongest resistance to acid-induced quenching deriving from thick insulating Cd_xZn_1−xS layers.

QD_17.4–SAv exhibited substantially more nonspecific binding to cells, with 218 QDs per GFP-negative cell, such that there was essentially no difference between GFP-positive and GFP-negative cells (p > 0.1, Figure 5f). This effect was not observed on glass coverslip surfaces (Figure S1) and did not correlate with the proportion of BSA adsorption across the three-QD series (Figure 2d), meaning that it was induced by specific interaction with cell surfaces. This high nonspecific binding of QD_17.4 is likely due to its flatter surface, for which a reduced curvature leads to higher contact area with biological surfaces. This is consistent with previous studies showing that between the size range of 14–50 nm, larger PEG-coated spherical nanoparticles are more efficiently associated with cells^41,42 and that flattened nanomaterials such as nanoplatelets exhibit greater nonspecific binding to membranes compared with spherical variants with the same coating.⁴³ We are continuing to study this effect to gain a greater understanding of the fundamental underlying mechanisms.

We evaluated alternative experimental conditions to determine how universal these results were. First’ we tested the impact of an alternative colloidal blocking agent, casein, which, in comparison with BSA, is more compatible with primary neurons, as applied below. Nonspecific binding of QD_17.4 on live GFP-negative HeLa was reduced 4-fold with casein (Figure S8); however, specific binding to GFP-positive HeLa cells also decreased proportionally, such that there was still no significant difference between the two groups. We further tested this three-QD series on a different cell type, HEK 293T, which is an immortalized embryonic kidney cell line that is ~3 times smaller by area compared to HeLa cells. Trends were very similar to that observed in HeLa cells, with QD_13.3 yielding the highest signal-to-background ratio and QD_17.4 exhibiting the most nonspecific binding (which was lower than in HeLa cells), resulting in no significant difference between GFP-positive and GFP-negative cells (Figure S9).

Due to the much higher nonspecific binding of QD_17.4 compared with QD_9.2 and QD_13.3, we designed, synthesized, and tested an additional QD with size intermediate between QD_13.3 and QD_17.4. We prepared a quasi-spherical (core)shell (CdSe)Cd_xZn_1−xS QD with 7.29 ± 0.82 nm TEM diameter and 15.2 nm hydrodynamic diameter measured by GPC after coating with P-IM-N₃ (Figure S10). This QD, with size midway between QD_13.3 and QD_17.4, is denoted as QD_15.2. QD_15.2–SAv conjugates yielded 47 nonspecifically bound QDs on live GFP-negative HeLa cells, which was much greater than that of QD_13.3–SAv (<1) but substantially smaller than that of QD_17.4–SAv (218) (Figure S11). The mean number of QD_15.2–SAv bound to GFP-positive cells was larger than that for GFP-negative cells; however, like QD_17.4–SAv, there was no statistical difference between the control and target cells (p > 0.1) (Figure S11). These results indicate that there is an abrupt enhancement in nonspecific binding for QDs with cores larger than 5.5 nm and coated with OEG with three repeating units. Longer OEG chains, and much longer PEG polymers, will further resist nonspecific binding effects due to additional conformational entropy of the coating but at the expense of an increase in size.⁴⁴

Labeling Receptors in Fixed and Permeabilized HeLa Cells.

We next evaluated the impact of QD size on quantitative labeling of receptors in cells that are fixed and permeabilized instead of live, a common biological labeling scenario for QDs. We used the same B-GluR2 expression system and again used HeLa cells (Figure 6a). B-GluR2, like most membrane receptors, is present on both the membrane and within the cytoplasm, which results in larger number of accessible targets when the cells are permeabilized. The cells were imaged in widefield epifluorescence mode, revealing QDs that were distributed much more densely throughout fixed cells (Figure 6b) compared to live cells (Figure 5e). For all three QD sizes, staining in GFP-positive cells was significantly greater than that in GFP-negative cells (Figure 6c; p < 0.005), even for QD_17.4–SAv, for which there was no dependence on GFP expression in live cells (Figure 5f). To compare staining across the three QD sizes under this dense labeling scenario in which individual QDs could not be discretely counted, we normalized the integrated intensity per cell by the relative QD brightness values at the single-molecule level measured on glass coverslips to calculate the number of QDs bound (Figure 4d). For QD_13.3, we calculated 10,743 specific and 4,186 nonspecific counts on average, which is much larger than that observed in live cells (Figure 5f) due to the thicker illumination plane in epi-illumination mode and the comprehensive access of probes to targets beyond the membrane in intracellular compartments. Consistent with the case of live cells, nonspecific binding of QD_17.4–SAv was greater than that of QD_13.3–SAv, which exhibited a high signal-to-background ratio that was similar to that of a Alexa594-conjugated SAv, which we consider to be the gold-standard label (Figure S13). However, unlike in the case of live cells, the smallest QD_9.2–SAv surprisingly yielded the highest number of QDs nonspecifically bound to fixed cells, resulting in a poor signal-to-background ratio that was similar to that of QD_17.4–SAv (Figure 6d).

Figure 6.

Evaluation of membrane receptor labeling density and specificity using three QD sizes in fixed and permeabilized HeLa cells. (a) Schematic showing the workflow used for HeLa cell transfection with the same plasmid in Figure 5a, followed by fixation, permeabilization, and staining with QD–SAv. (b) Representative images of QDs binding to GFP-positive cells expressing B-GluR2 or GFP-negative cells after 30 min treatment with 10 nM QD–SAv at room temperature. All images show the same image acquisition conditions except for the red-outlined inset for QD_17.4, for which intensity is reduced to observe individual QDs. Nuclei are shown in blue; GFP is shown in green; scale bar = 20 μm. (c) Quantification of number of QDs bound to individual GFP-positive and GFP-negative cells. Intensity is scaled to reflect QD number calculated by dividing QD fluorescence per cells by the average single QD intensity shown in Figure 4d. A.U. = arbitrary units. (d) Ratio of number of QDs bound to GFP-positive cells divide by QDs bound to GFP-negative cells from panel (f). Bar graph indicates mean ± s.e.m. (e) Fraction of QDs in nucleus of cells GFP-negative cells. N = 29, 47, and 32 for QD_9.2, QD_13.3, and QD_17.4, respectively. In (c) and (d), for GFP-negative cells, N = 29, 47, and 32 for QD_9.2, QD_13.3, and QD_17.4, respectively. For GFP-positive cells, N = 31, 41, and 28 for QD_9.2, QD_13.3, and QD_17.4, respectively. In (c) and (e), box plot shows 25/75 percentile; red lines indicate mean values; whiskers indicate s.d. Note that the difference in intensity detected in QD channels between GFP-positive and GFP-negative cells is solely due to QD signals and not due to GFP signal crosstalk (Figure S12). In (c) and (e), ** indicates p < 0.005.

The nonlinear trend in intracellular nonspecific binding with QD size can be explained by known biophysical processes governing colloidal interactions with fixed cells. Compared with intact living cells, fixation results in the exposure of diverse microenvironments containing hydrophobic residues and charged domains from denatured proteins that can adsorb OEG- and PEG-coated colloids.^45-48 Furthermore, unlike live cells in which the membrane is the only accessible structure, permeabilization provides access to the cell interior, which is a crowded mesh of macromolecules. Experimental measurements and theoretical models have shown that the cytoplasmic pore size is polydisperse with a mean near 30 nm in living cells,⁴⁹ and becomes more restrictive upon protein crosslinking induced by fixation.^50,51 All of the QD–SAv conjugates used in this work are within a size range permitting permeation into the cytoplasm, however more microdomains are expected to be available for smaller QDs. Therefore, while the greater nonspecific binding of QD_17.4–SAv compared with QD_13.3–SAv can be explained by the high adsorption affinity of QD_17.4–SAv previously observed in live cells (Figure 5f), the highest nonspecific binding of QD_9.2–SAv can be explained by its more extensive steric access to sticky intracellular domains. This is consistent with trends apparent in Figure 6b,e showing there is a progressive accumulation of QDs in the nucleus with decreasing nanocrystal diameter in GFP-negative cells. Trends were similar in GFP-positive cells but with smaller differences between the probe sizes (Figure S13). Compared with the cytoplasm, the nuclear pore is much more restrictive to passive transport of colloids, limiting diffusion for 5 nm materials in the live-cell state.⁵² Therefore, the smallest QD_9.2 would be expected to enter the nucleus and other microdomains to the greatest degree, whereas QD_17.4 would exhibit the most restricted access. Notably the compartmental staining pattern of Alexa594-labeled SAv was most similar to that of QD_17.4–SAv (Figure S13). This indicates that enhanced nuclear labeling with smaller QDs was not due to the presence of biotinylated target in the nucleus, but simply due to nonspecific binding. This effect is consistent with previous reports demonstrating a size-dependent exclusion of nanomaterials from the nucleus and the nonspecific affinity of small QDs to nuclear targets.^53,54

Detailed molecular processes governing nonspecific binding are not clear due to the lack of understanding of specific functional moieties and cell structures involved in adsorption in fixed cells.^55-57 However, it is clear that washing cannot readily desorb the QDs, as the extensive wash steps implemented here with saline solutions did not lead to free diffusing QDs in the incubation medium. The best way to further eliminate nonspecific binding, therefore, is to prevent the initial adsorption process. We have found through wide screening of solution conditions and preliminary incubation steps that highly specific intracellular labeling (signal-to-background ratio >20) of diverse low-copy number nucleic acid targets can be achieved with QD probes if the cells are initially incubated with a combination of a colloidal blocking agent (BSA) to adsorb hydrophobic domains and an anionic blocking agent (dextran sulfate) to adsorb to polycationic sites.⁸ The success of this procedure leads us to believe that intracellular polycationic domains are the primary mediator of nonspecific binding of QDs, however, unfortunately this mixture is not compatible with intracellular protein labeling as polyanions disrupt protein structures (data not shown). Further progress can likely be achieved by exploiting recent developments in zwitterionic nanomaterial coatings which can repel polycationic and hydrophobic domains.^45,58,59

Labeling Receptors in Live Hippocampal Neuron Synapses.

We next evaluated the capacity of these three QD sizes to label the GluR2 receptor subunit in its orthotopic location in the synaptic cleft of living hippocampal neurons. The synaptic cleft is a ~20–30 nm wide space between interconnected neurons where intercellular communication is mediated through neurotransmitters and their cognate receptors (Figure 7a,b).⁶⁰ Previous studies have shown that larger labels cannot access molecular targets within the synapse due to size exclusion.^19,20 Here, we cotransfected primary rat hippocampal neurons with three plasmids containing GluR2-AviTag, BirA, and Homer1c-mGeos. BirA biotinylates the AviTag on GluR2, enabling detection of GluR2 in synapses by QD–SAv labels. Homeric localizes specifically on the postsynaptic dendrites so that the fused photoswitchable fluorescent protein, mGeos, can be used to resolve the position of a synapse on a dendrite at super-resolution using photoactivation localization microscopy (PALM).^19,20 In live hippocampal neurons, we applied QD–SAv probes at a 100-fold lower concentration of QD–SAv (100 pM) and 24-fold shorter incubation time (5 min) compared to the conditions used for live HeLa cells, together with casein-based blocking. These conditions allow sparse labeling of B-GluR2 for single-molecule detection such that each synapse has at most one QD (Figure 7c). Under these conditions, there was little nonspecific binding to GFP-negative cells, with fewer than 1 QD bound per 1000 μm² for all three QD sizes. We defined QD-GluR2 spots as being synaptic if they were within 0.5 μm of Homer1c, a criterion from previous studies.^19,20 We found that a reduction QD–SAv size led to a progressive increase in the percentage of QDs in synapses (Figure 7c,d), a trend consistent with previous results.^19,20 Specifically, 56% of QD_9.2–SAv was synaptic, which decreased to 41% for QD_13.3–SAv, and to 31% for QD_17.4–SAv. The upper limit of synaptic access is likely to be 77% based on the same measurements performed using SAv labeled with an organic dye, totaling ~5 nm in hydrodynamic size.¹⁹ This demonstrates a progressive benefit to using smaller probes for synaptic labeling to minimize measurement biases. In addition, these results show that conditions used for dense molecular counting and conditions used for sparse labeling for single-molecule tracking can have substantially different results and limitations even when applying identical probes. Therefore, to achieve the desired degree of target saturation and minimize the impact of nonspecific binding with recently developed classes of small QDs, multiple conditions should be optimized on the biospecimen under study, including the probe concentration, labeling time, and blocking agents.

Figure 7.

Evaluation of membrane receptor labeling specificity using three QD sizes in synapses of live primary hippocampal neurons. (a) Schematic depiction of synapse between two neurons. (b) Schematic depiction of the synapse showing the narrow cleft (~20 nm separation), the B-GluR2 plasma membrane receptor, and the QD–SAv label. (c) Representative fluorescence micrographs (left) and super-resolution reconstructed images of the white outlined box (right) showing neurons with Homer1c-mGeos contrast in green labeling postsynaptic clusters to identify synapses, with QD–SAv labels for B-GluR2 shown in red. Images are shown for each of the three QD sizes. Scale bar = 5 μm. (d) Distances between QD–SAv spots and the centroid of postsynaptic clusters from Homer1c labels. Green boxes show the fraction of QD–SAv within the synapse, defined by a minimum separation distance of 0.5 μm between the QD centroid and the postsynaptic clusters.

CONCLUSIONS

In recent years, QD probes have significantly influenced single-molecule imaging, tracking, and counting, providing an unmatched combination of fluorescence intensity and signal duration in a compact label.^3-6 The recognition that the finite size of these colloidal nanocrystals contributes to measurement biases has led to a concerted effort by many laboratories to further reduce the size of these materials,^23,61-68 which is also expected to benefit other applications in tissue staining⁸ and in vivo imaging.^69,70 Current designs can now reach 7–9 nm, which is markedly smaller than the standard size of ~25–35 nm a decade ago.²² This improvement was driven largely by advances in smaller surface coatings, rather than developments in nanocrystals. The smallest hydrodynamic sizes require application of the smallest (core)shell nanocrystals, however, as this study shows, photophysical benefits of QDs can be lost when the nanocrystal is reduced to a certain size. In particular, a reduced brightness and greater off-time blinking fraction led to less satisfactory molecular counting results for QD_9.2 compared with QD_13.3 when applied for comprehensive target labeling. Nevertheless, the smallest QD_9.2 was still the superior probe for dynamic tracking of targets in cell regions with size cutoffs for entry such as synapses. We emphasize that this work focused on one molecular target (GluR2 membrane receptors) using a single bioaffinity tag (streptavidin) in order to focus singularly on the impact of nanocrystal size. However, we have applied variants of these QDs with the same coatings to label diverse mRNA and protein species in cells and tissues and found that the absolute signal-to-background ratio depends on both the target and biospecimen, with values of ~6 for cytoplasmic inducible nitric oxide synthase protein,⁵⁸ > 20 for cytoplasmic mRNA sequences,⁸ and >100 for epidermal growth factor receptor in live cells.⁷

There is clearly more work needed to maximize the utility of the smallest QDs. The smallest QD labels are still larger than widely applied GFP labels (~4 nm) and organic dyes (~1 nm).³ While considered an authoritative standard for protein localization when validation controls are satisfied, the 4 nm size of GFP is a reasonable goal for the community to further realize benefits in biological measurements. Approaching this goal will require further developments in thin coatings as well as ultrasmall QD nanocrystals. The 3.2 nm (core)shell alloyed nanocrystal used to generate QD_9.2 is the best that we have produced to date to minimize nanocrystal size while retaining high QY in water, photostability, and red-shifted emission and excitation. Both absorption and brightness could be boosted by using a larger core compensated by a thinner shell, but at the cost of a decrease in QY and increase in blinking. On the other hand, a smaller core with thicker shell will increase QY and decrease blinking, but the brightness will be lower. It is possible that alternative material families with intrinsically higher absorption coefficients may be needed to increase brightness of the smallest nanocrystal sizes; however, no alternative compositions to date have matched the combination of wavelength tunability, high QY, and stability for the II–VI family used herein. We anticipate that the materials characterizations and systematic performance evaluations reported in this study can provide guidance to selecting appropriate compact QDs for molecular labeling applications and also inform the further design of smaller QDs.

METHODS

Materials.

2-Azidoacetic acid (97%), 1,2-hexadecanediol (HDD, 97%), behenic acid (BAc, 99%), cadmium acetate hydrate (Cd(Ac)₂·H₂O, 99.99+%), mercury acetate (Hg(Ac)₂, 99.999%), N-methyl-formamide (NMF, 99%), N,N,N′,N′-tetramethylethylenediamine (TEMED, 99%), 1-octanethiol (OT, ≥ 98.5%), selenium dioxide (SeO₂, ≥ 99.9%), selenium powder (Se, 99.99%), sulfur powder (S, 99.98%), tetramethylammonium hydroxide solution (TMAH, 25 wt % in methanol), dimethyl sulfoxide (DMSO), Tris hydrochloride (Tris-HCl, 1M), and casein powder were purchased from Sigma-Aldrich. Zinc acetate (Zn(Ac)₂, 99.98%) and cadmium chloride (CdCl₂, 99.99%) were acquired from Alfa Aesar. 1-Octadecene (ODE, 90%), oleylamine (OLA, 80–90%), and oleic acid (OAc, 90%) were from Acros Organics. DBCO-sulfo-NHS ester (95%) was from Click Chemistry Tools. Organic solvents including acetone, chloroform, diethyl ether, hexane, and methanol were purchased from multiple sources, including Acros Organics, Fisher Scientific, and Macron Fine Chemicals. Dulbecco’s Modified Eagle’s Medium (DMEM), fetal bovine serum (FBS), Hoechst, penicillin-streptomycin stock solutions (P/S), PBS, Triton-X 100, trypsin EDTA, l-glutamine, sodium pyruvate, and Lipofectamine 2000 were acquired from Fisher Scientific. Cell culture grade BSA was from GE Healthcare. SAv was from ProSpec. Paraformaldehyde (PFA, 32% v/v in water) was from Electron Microscopy Sciences. Biotinylated DNA (5′-Biotin/(T)₆₈ TAG CCA GTG TAT CGC AAT GAC G-3′) was from Integrated DNA Technologies. Casein was from Vector Laboratories. Cadmium behenate (Cd(BAc)₂), cadmium myristate (Cd(MAc)₂), and mercury octanethiolate (Hg(OT)₂) were synthesized following methods in our previous manuscript.¹¹

QD_13.3 Nanocrystal Synthesis.

CdSe cores were synthesized following literature methods.⁷¹ Cd(BAc)₂ (0.2 mmol), SeO₂ (0.2 mmol), HDD (0.2 mmol), and ODE (5 mL) were added to a 25 mL round-bottom flask that was then degassed at ~120 °C for 60 min. Under a nitrogen atmosphere, the temperature was ramped to 240 °C at ~20 °C/min, and QD growth was allowed to occur at 240 °C for 60 min. The mixture was cooled to ~100 °C and OAc (1 mL) was injected into the flask, which was then cooled to room temperature. The reaction mixture was transferred to a 50 mL centrifuge tube and the CdSe cores were precipitated by the addition of methanol (5 mL) and acetone (25 mL). The QDs were dispersed in 5 mL of hexane and precipitated again with methanol and acetone. The QD precipitate was washed with acetone to completely remove methanol and then dispersed in hexane. The purified CdSe cores had a first exciton absorption band at 538 nm and an emission band at 549 nm. Then a Cd_yZn_1−yS shell was deposited layer-by-layer by successive addition of cationic and anionic precursors following literature methods with some modifications.⁷² The purified CdSe cores (~300 nmol) dispersed in hexane (5 mL) were transferred to a 50 mL round-bottom flask together with ODE (4 mL) and OLA (2 mL). The mixture was evacuated at 50 °C until the hexane was fully evaporated. Under a nitrogen atmosphere, the reaction temperature was then increased to 120 °C and shell growth was initiated in 0.8 monolayer (ML) increments using sulfur, cadmium acetate, and zinc acetate precursors. Shell growth began with the sulfur precursor, which was added dropwise in 3–5 min, and allowed to react for 15 min. Next, the same amount of cadmium precursor was similarly added to complete the first cycle for a 0.8 ML shell. The reaction temperature was raised stepwise by ~10 °C between each precursor addition until reaching a maximum of ~190 °C and an aliquot (50 μL) was withdrawn and diluted 10-fold in hexane to measure the absorption and emission spectra after each cycle of growth. In total, the shell was composed of 2.4 ML CdS, 0.8 ML Cd_0.8Zn_0.2S, and 1.5 ML ZnS, yielding an emission band at 608 nm. After growth, the reaction was cooled to room temperature and the QDs were purified by precipitation with methanol and acetone, and redispersed in hexane for storage.

QD_17.4 Nanocrystal Synthesis.

QD_17.4 was synthesized in a manner similar to that used for QD_13.3 with the following modifications: (1) CdSe synthesis used Cd(MAc)₂ instead of Cd(BAc)₂ to generate cores with first exciton band at 645 nm and emission at 667 nm, (2) the shell growth was performed with 80 nmol of cores, (3) the shell growth was initiated at 160 °C instead of 120 °C, (4) the final shell growth temperature was 200 °C instead of 190 °C, and (5) the shell composition was 4.0 ML CdS, 0.8 ML Cd_0.5Zn_0.5S, and 1.6 ML ZnS. The emission band after shell growth was centered at 685 nm.

QD_9.2 Nanocrystal Synthesis.

The synthesis of CdSe cores for QD_9.2 was similar to that of QD_13.3 with the following modifications: (1) 0.05 mmol of SeO₂ and 10 mL of ODE were used in the reaction, (2) the maximum reaction temperature was 230 °C, and (3) once the reaction solution changed from colorless to faint yellow, the reaction was quenched by removing the heating mantle. After purification of the QD cores and dispersion in hexane, the solution was centrifuged at 3000g for 5 min to remove unreacted cadmium behenate. The purified CdSe cores had a first exciton absorption band at 488 nm and an emission band at 500 nm. The cores were then exchanged with mercury to redshift the optical spectra. CdSe cores (~100 nmol) in hexane were mixed with OLA (4 mL) and ODE (l mL) under nitrogen, and hexane was evaporated under vacuum at 50 °C. Mercury exchange was initiated with the injection of Hg(OT)₂ (0.05 mmol) dissolved in OLA (l mL). The reaction was monitored by removing aliquots and measuring the absorption spectrum while gradually increasing of the temperature to 110 °C. Once the desired extent of redshift was achieved, the reaction was quenched by removing the heating mantle. The Hg_xCd_1−xSe cores were precipitated with methanol and acetone and then dispersed in hexane (10 mL). The QD dispersion was centrifuged at 3000g for 5 min to remove unreacted mercury precursor before the cores were washed two more times with methanol and acetone, and finally redispersed in hexane. The purified cores had a first exciton absorption band at 530 nm and emission band at 618 nm. Shell growth for QD_9.2 was performed in a similar manner to that of QD_13.3 with the following modifications: (1) 75 nmol of Hg_xCd_1−xSe cores were used, (2) the solvent was composed of 2 mL of ODE and 1 mL of OLA in a 25 mL round-bottom flask, (3) shell growth was initiated at 110 °C, and (4) the shell composition was 1.6 ML Cd_0.5Zn_0.5S and 0.8 ML ZnS. After addition of all precursors, the reaction temperature was raised to 175 °C at a rate of ~20 °C/min before the reaction was quenched by removing the heating mantle.

QD_15.2 Nanocrystal Synthesis.

QD_15.2 was synthesized using a method similar to that of QD_13.3 with the following modifications. The CdSe core was synthesized using Cd(MAc)₂ instead of Cd(BAc)₂, yielding a first-exciton absorption band at a wavelength of 604 nm. Layer-by-layer shell growth was carried out with ~30 nmol of CdSe core with an initial temperature of 160 °C instead of 120 °C. The final temperature was 220 °C instead of 190 °C. The shell was composed of 4.8 ML CdS and 1.6 ML ZnS.

Polymer Coating of QDs.

The QDs were coated with P-IM-N₃ by following previously reported methods with slight modifications.²³ After synthesis, 0.5 mL of hexane dispersions of QD_9.2 (20.50 μM), QD_13.3 (18.3 μM), or QD_17.4 (3.92 μM) were diluted to 1.5 mL with hexane and precipitated with a mixture of methanol (2 mL) and acetone (10 mL) before redispersion in hexane (3 mL). These QDs (3.1 μM for QD_9.2, 2.5 μM for QD_13.3, 0.14 μM for QD_17.4) were mixed with NMF (2 mL) and TMAH (292 μL for QD_9.2, 250 μL for QD_13.3, 57 μL for QD_17.4) in a 7 mL glass vial equipped with stir bar. After 1 h of stirring, QDs were fully transferred into NMF. The hexane phase was removed, and excess methanol and hexane were evaporated under vacuum for 1 h. Each QD in NMF (1 nmol for QD_9.2 and QD_13.3, 0.1 nmol for QD_17.4) was diluted in DMSO (750 μL) in a clean 7 mL glass vial equipped with a stir bar. P-IM-N₃ (11.3 mg/mL, 56 μL for QD_9.2, 169 μL for QD_13.3, 84 μL for QD_17.4) was then added dropwise to each QD dispersion. Polymer coating was carried out at 110 °C with stirring for 2 h. After 2 h, QDs were precipitated with a mixture of ether (5 mL) and chloroform (2 mL). The QD pellet was then dispersed in 50 mM sodium borate buffer (pH 8.5). Excess polymer was removed using a 30 kDa molecular weight cutoff (MWCO) centrifuge filter (Amicon).

QD Conjugation to SAv.

QDs coated with P-IM-N₃ were conjugated at a 1:1 molar ratio to DBCO-modified SAv following our previous report.⁷ To verify functionality of the conjugates, QD–SAv was reacted with biotinylated DNA (B-DNA) for 4 h at 4 °C in PBS. The products were characterized by electrophoresis in a mixed polyacrylamide-agarose gel (2% polyacrylamide and 0.5% agarose).

QY and Brightness Measurements.

After polymer coating, QD dispersions were exchanged with PBS using 50 kDa MWCO Amicon centrifuge filters. The QY of each QD was measured using fluorescein in 0.1 M NaOH (QY = 92%) as a reference using methods in our previous publication.¹¹ Brightness was calculated using eq 1 and the known molar extinction coefficients from the QDs determined during synthesis.

FCS Measurements and Size Calculations.

BSA solutions (1% w/v) were prepared in PBS and cooled to 4 °C. QD stocks were diluted in 0% or 1% BSA solution to a final concentration of 10 nM. Samples were then transferred to eight-well glass-bottom Lab-Tek chambers (Thermo Scientific) and recorded using an Alba FCS instrument (ISS) with diode laser (470 nm) for excitation and single-photon avalanche photodiode detector. Fluorescence time-traces were acquired for 10 s at a frequency of 100 kHz. The confocal spot size of the FCS instrument was measured using a dye standard with a known diffusion coefficient (rhodamine B, D = 4.2 × 10⁻¹⁰ m² s⁻¹)⁷³ using the equation below⁷⁴

$$G(\tau )=\frac{1}{N}{\left(1+\frac{\tau }{{\tau }_{\mathrm{D}}}\right)}^{-1}{\left[1+\frac{\tau }{{\tau }_{\mathrm{D}}}{\left(\frac{{\omega }_{xy}}{{\omega }_z}\right)}^2\right]}^{-1∕2}$$ (2)

where G(τ) is the autocorrelation function, N is the average number of dye molecules in the confocal volume, τ_D is the diffusion time related to D according to eq 3, and ω_xy and ω_z are the xy-radius and z-radius of the confocal spot, respectively.

$$D=\frac{{\omega }_{xy}^2}{4{\tau }_{\mathrm{D}}}$$ (3)

Fitting of eq 2 to the data for rhodamine B yielded ω_xy = 0.299 μm and ω_z = 3.68 μm. These values were then used to fit the diffusion time τ_D for QDs using the equation below⁷⁵

$$G(\tau )=\frac{1}{N}[1+\theta e^{(\frac{\tau }{{\tau }_{\text{blink}}})}]{\left(1+\frac{\tau }{{\tau }_{\mathrm{D}}}\right)}^{-1}{\left[1+\frac{\tau }{{\tau }_{\mathrm{D}}}{\left(\frac{{\omega }_{xy}}{{\omega }_z}\right)}^2\right]}^{-1∕2}$$ (4)

where θ accounts for QDs in nonfluorescent states and τ_blink is the characteristic blinking time for QDs described previously.⁷⁶ The diffusion coefficient of each QD was calculated using eq 3 above, and the hydrodynamic diameter (HD) was calculated using the Stokes–Einstein equation

$$\text{HD}=\frac{k_{\mathrm{B}}T}{3\pi \eta D}$$ (5)

where k_B is the Boltzmann constant, T is the temperature (20 °C or 293.15K), and η is the viscosity of the solution. The viscosity of 1% BSA in PBS was calculated using a linear approximation reported previously⁷⁷ with an intrinsic viscosity of BSA of 3.7 cm g⁻¹.⁷⁸

QD Sensitivity to pH.

QDs (20 nM, 100 μL) were incubated in Briton-Robinson buffer (40 mM) at different pH (4, 5, 6 and 7.2) in a black 96 well-plate at room temperature. The fluorescence signals were recorded every 30 min starting at 1 h using a SpectraMax M2 microplate reader. At each time point, fluorescence signal intensity was normalized to the average intensity at pH 7.2.

Plasmids and Cloning.

The AviTag-GluR2/eGFP/BirA plasmid used in this study was created through modification of an AviTag-GluR2 plasmid. A gBlock Gene Fragment (Integrated DNA Technologies) encoding eGFP and a T2A peptide under control of an EF1a core promoter (EFS) was ligated into the backbone through Gibson assembly. A DNA fragment encoding an IgK leader followed by BirA was generated by PCR from pDisplay-BirA-ER (Addgene plasmid #20856) and cloned in frame with eGFP and the T2A peptide through Gibson assembly. The plasmid sequence is included in the Supporting Methods.

Fluorescence Microscopy.

Fluorescence images of live and fixed HeLa cells and QDs on coverglass were acquired via wide-field illumination on a Zeiss Axio Observer Z1 inverted microscope equipped with a 100× 1.45NA alpha Plan-Fluar oil immersion objective with 488 nm/100 mW OPSL laser and 561 nm/40 mW diode laser units, 100 W halogen lamp for illumination, and Photometrics eXcelon Evolve 512 EMCCD camera. For data acquisition, Zeiss ZEN software was used. Semrock and Zeiss filters (G 365, BP 470 nm/40 nm, BP 482/18, BP 561/14 nm) were used to filter excitation light, and Semrock filters (BP 609/54 nm, BP 650/100 nm, BP 698/70 nm, and LP 585) were used to filter emission light. Brightfield images were acquired with a 12 V, 100 W Halogen lamp through a DIC prism III/0.55. Images were exported as 8-bit 512 × 512 TIFF files. Fluorescence images of neurons were acquired using a Nikon Eclipse Ti microscope equipped with a Nikon APO 100× objective (1.4 NA). The sample was stabilized in the z-axis using a Perfect Focus System (PFS). The sample was illuminated with an Agilent laser system (MLC400B) with four fiber coupled lasers (405 nm, 488 nm, 561 nm, 640 nm). Images were recorded using a back illuminated EMCCD camera (Andor DU897). A cylindrical lens (CVI Melles Griot, SCX-25.4-5000.0-C-425-675) of focal length 10 m was inserted below the back aperture of the objective for 3D imaging. For x-y-z stage control, a motorized stage with a piezo top plate (ASI PZ-2000FT) was used. Nikon Elements software was used for data acquisition. For imaging, BP 655 nm/50 nm, BP 600 nm/50 nm, BP 710 nm/40 nm were used. All quantum dots were imaged using 561 nm laser illumination. For imaging Homer1c-mGeos, a band-pass emission filter 525/50 and 488 nm laser illumination were used. A quad-band dichroic (Chroma, ZT 405-488-561-640RPC) was used for separating fluorescence emission from illumination. A separate camera (The Imaging Source LLC, DMK 23U74) was used for simultaneously imaging beads for drift correction.

Imaging and Intensity Analysis of Single QDs.

QDs coated with P-IM-N₃ in PBS (100 pM, 500 μL) were incubated in 4-well glass bottom CELLview dishes (Gibco) for 2 min at room temperature. Excess QDs were washed away with PBS (3 times) to leave QDs absorbed sparely on the glass surface. QDs were excited with a 561 nm laser at 80% power and 100 ms exposure time for 900 frames (QD_9.2 and QD_13.3) or 4000 frames (QD_17.4). Images were exported as 8-bit TIFFs. Time-course images of the same field of view were combined as a single TIFF stack, imported into MATLAB, and processed using a custom code for single QD identification and intensity analysis as reported previously.^7,11 Briefly, the positions of QD spots [x₀,y₀] were detected using the detection/estimation/deflation algorithm from the multiple-target tracing (MTT) algorithm.³⁹ The intensity of 3 × 3 pixels centered around [x₀,y₀] was calculated. The time-course intensity of each QD was binned into a histogram with 100 bins which was fit to a sum of a Gaussian background with integrated area a₁, and a skewed Gaussian QD signal with integrated area a₂. Single QDs were identified as those satisfying the single molecule behavior of a single QD, with (1) each QD in the “on” blinking state for at least 5% of frames, (2) each QD in the “on” for no more than 95% of frames, (3) relative variance for all fitting parameters less than 500%, and (4) a correlation coefficient greater than or equal to 0.98 between fit and data. The single-QD intensity was defined as the average intensity across the time course. The off fraction was calculated as

$$\text{off fraction}=\frac{a_1}{a_1+a_2}$$ (6)

Treatment of HeLa Cells with QD–SAv and Alexa594–SAv.

HeLa cells (ATCC CCL-2) were cultured in DMEM supplemented with 10% FBS and 1% P/S. When they reached 70% confluence, cells were harvested and seeded (300 μL, 11000 cells/mL) in each well of an eight-well Lab-Tek chamber. After 24 h, cells reached 90% confluence and were transfected with the plasmid containing AviTag-GluR2, eGFP, and BirA sequences (0.6 μg). After transfection for 4 h, the medium was changed to phenol red-free DMEM supplemented with 10% FBS, 584 mg/L l-glutamine, and 110 mg/L sodium pyruvate. After 48 h, cells were used for either live- or fixed-cell experiments. For live-cell imaging, the medium was changed to serum-free, phenol red-free DMEM supplemented with 0.5% w/v BSA, 584 mg/L l-glutamine, and 110 mg/L sodium pyruvate (imaging media). Cells were then stained with Hoechst (1 μg/mL) for 20 min at 37 °C in imaging media. Excess Hoechst was washed off with imaging media and cells were blocked with either 0.5% w/v BSA or 0.5% w/v casein in serum-free, phenol red-free DMEM supplemented with 584 mg/L l-glutamine and 110 mg/L sodium pyruvate for 15 min at 37 °C. QD–SAv and Alexa594–SAv (ThermoFisher) were diluted to 10 nM in imaging media at room temperature, added to the cells, and incubated for 30, 60, or 120 min at 37 °C. Unbound probes were then removed by 3 washes with imaging media. For fixed cells, cells were treated with 4% PFA for 15 min, permeabilized with 0.1% Triton-X at room temperature for 15 min, and blocked with 1% w/v BSA in PBS for 1 h at room temperature. QD–SAv and Alexa594–SAv conjugates were diluted to 10 nM in 1% w/v BSA in PBS at room temperature and added to the cells for 30 min at room temperature. Unbound probes were removed by two washes with 1% BSA in PBS and one wash with PBS. Cells were stained with Hoechst (1 μg/mL) for 10 min at room temperature before three washes with PBS.

Treatment of HEK 293T Cells with QD–SAv.

HEK 293T cells (ATCC CRL-3216) were cultured in DMEM supplemented with 10% FBS and 1% P/S in 12-well plates. When reaching 50% confluence, cells were transfected with the plasmid containing AviTag-GluR2, eGFP, and BirA sequences (3 μg). After 24 h of transfection, cells (300 μL, 16000 cells/mL) were harvested and seeded in an 8-well Lab-Tek chamber. After 24 h, the medium was changed to phenol red-free DMEM supplemented with 10% FBS, 584 mg/L l-glutamine, and 110 mg/L sodium pyruvate. After 48 h, cells were used for live-cell imaging using conditions and workflows described above for HeLa cells.

Imaging and Analysis of QD–SAv and Alexa594–SAv on HeLa Cells.

For live cells, all images in brightfield, Hoechst fluorescence, GFP fluorescence, and QD and Alexa594 fluorescence channels were acquired in a single z-plane near the center of the cell nuclei. Nuclear and GFP images were acquired in epifluorescence mode whereas QD signals were acquired using the HILO method.⁷⁹ Because AviTag-GluR2, GFP, and BirA are on the same plasmid, HeLa cell expression of B-GluR2 was identified based in the presence or absence of GFP signals. Individual cells were manually segmented based on their outlines visible in brightfield images. QD spots in each cell were detected and counted using the detection/estimation/deflation algorithm from the MTT algorithm³⁹ in MATLAB.

For fixed cells, all images were acquired in a single z-plane near the center of cell nuclei. Nuclear, GFP, Alexa594, and QD images were acquired in epifluorescence mode. B-GluR2 positive and negative cells were identified and segmented as described above for live cells. Nuclei were manually segmented based on Hoechst channels. QD, Alexa594, and GFP intensity per cell or per nucleus was calculated using a custom MATLAB code. QD intensity per cell (I_QD,Cell) was calculated by subtracting the mean background autofluorescence (AF) as

$$I_{\text{QD,cell}}=I_{\text{total,cell}}-{\overline{I}}_{\text{AF,A}}\times A_{\text{cell}}$$ (7)

where I_total,Cell the is summed intensity per cell, A_cell is the cell area in pixels, and ${\overline{I}}_{\text{AF,A}}$ is the mean AF intensity per pixel for cells measured without QDs added.

Culture and QD–SAv Treatment of Neurons.

Primary hippocampal neurons were isolated from E18 rat embryos using methods previously described,¹⁹ approved by the Institutional Animal Care and use Committee at the University of Illinois. Cells were seeded on 25 mm coverslips coated with poly-l-lysine. Carboxyl beads (1 μm, Bangs Laboratories, #PC04N) were sparsely coated on the coverslips for use for stage drift correction. At 13 days in vitro (DIV), neurons were cotransfected with plasmids encoding Homer1c-mGeos (1 μg/mL), GluA2-AP (1 μg/mL), and ER-retained biotin ligase (BirA-ER, 1 μg/mL) using Lipofectamine-2000 transfection reagent in order to label the postsynaptic density and biotinylate the GluA2 subunit of AMPA receptors. After 48 h, the coverslips were transferred to warm HEPES-buffered saline (HBS), containing 140 mM NaCl, 5 mM KCl, 25 mM HEPES, 10 mM D-Glucose, 1.5 mM MgCl₂ and 2 mM CaCl₂ at pH 7.4 and incubated for 5 min. The coverslips were then transferred to an imaging chamber (Warner RC-40LP) and incubated for 5 min with QD–SAv (200 μL, 100 pM) in warm HBS containing 2.5% casein to sparsely label the B-GluA2 subunit of AMPA receptors on the cell surface. Excess QDs were washed off using warm HBS. The Nikon Eclipse Ti microscope described above was used to image neurons. After focusing the sample in brightfield mode and finding a transfected cell using the GFP channel, AMPA receptors labeled with QDs were imaged and tracked for 1000 frames with 50 ms exposure time and 100 EM gain in HILO fluorescence mode. For PALM imaging,⁸⁰ mGeos was imaged for 2000 frames with 50 ms exposure time and 100 EM gain in HILO fluorescence mode. A 405 nm laser pulse for 100 ms was provided every 20 frames to activate mGeos proteins from the dark to green fluorescent state. Z-calibration for 3D imaging using the cylindrical lens astigmatism was performed as described previously⁸¹ and applied to both the PALM and single particle data.

Analysis of Synaptic Labeling.

Stage drift correction and chromatic aberration correction were performed as described previously,¹⁹ the only difference being that carboxyl beads were used instead of fiduciary markers. Trajectories of single QD–SAv were obtained using a previously described code.¹⁹ To identify synapses, the positions of detected proteins in each frame of the PALM data were localized and cluster analysis was performed. For each of the QD–SAv trajectories, the average distance of every point from the center of the nearest synapse was calculated. VMD 1.9.3⁸² was used to visualize the postsynaptic density clusters and QD–SAv trajectories.

Analysis of Nonspecific Binding in Neurons.

Cell areas that did not express Homer1c-mGeos signals were manually identified. QD spots per 1,000 μm² of these identified areas were counted using the detection/estimation/deflation algorithm from the MTT algorithm in MATLAB.³⁹

Binding of QD–SAv to Biotinylated Coverglass.

Glass surfaces were prepared using 50 well chambered coverglass (Electron Microscopy Sciences, 70460-50R) using our previously reported protocol.⁸³ Coverglass was cleaned using 1 M NaOH with sonication for 10 min, washed 10 times with water, and dried with a nitrogen flush. Surfaces were plasma treated and immediately incubated in a solution of methanol (93.46%), glacial acetic acid (4.67%) and 3-aminopropyltriethoxysilane (1.87%) for 10 min at room temperature. These amine-functionalized surfaces were then sonicated for 1 min, incubated for an additional 10 min at room temperature, washed 10 times with water, and dried with a nitrogen flush. A freshly prepared solution containing mPEG5000-NHS (2.375% w/v) or biotin-PEG5000-NHS (0.125% w/v) in sodium bicarbonate buffer (100 mM) was then applied (4 μL per well) and incubated in a humidified chamber for 3 h. The surfaces were then washed 10 times with water, dried by nitrogen flush, and stored at −20 °C until use. Prior to use, coverglass wells were blocked using a solution of 2% BSA for 1 h at room temperature in a humidified chamber. The surfaces were incubated with QD–SAv (10 nM) in PBS containing 2% BSA in the presence or absence of 10 μM biotin (4 μL per well) for 1 h at room temperature in a humidified chamber. The surfaces were then washed 6 times with 10 mL PBS, and imaged immediately. Samples were imaged via wide-field illumination using the aforementioned Zeiss Axio Observer.Z1 inverted microscope at 100× using a 488 nm laser for excitation at 100% power and a 585 nm long-pass filter, with 40 ms integration time and 10 EM gain.

Statistical Analysis.

Values are reported as mean ± standard deviation (s.d.) or mean ± standard error of mean (s.e.m.) as indicated. Statistical significance analyses were calculated using two-tailed student’s t-test in Origin Pro 9.1.

Instrumentation.

Absorption and fluorescence spectra of QDs were acquired using an Agilent Cary 5000 UV–Vis–NIR spectrophotometer and a Horiba NanoLog spectrofluorometer, respectively. For fluorescence and excitation spectra, wavelength-dependent detector sensitivity and excitation power fluctuations were accounted for during fluorescence signal measurement via a built-in module in the Horiba NanoLog spectrofluorometer. Electron microscopy images were collected using a JEOL 2100 Cryo TEM in the Frederick Seitz Materials Research Laboratory Central Research Facilities at the University of Illinois. Hydrodynamic sizes of polymer coated QDs were measured using an ÄKTApurifier UPC10 (GE Healthcare) with Superose 6 10/300GL column (GE Healthcare) and UNICORN 5.31 Workstation software. Fluorescence dependence on pH was measured using a SpectraMax M2 microplate reader at the Roy J. Carver Biotechnology Center at the University of Illinois. Gel electrophoresis was performed using an EPS-300X system (C.B.S. Scientific Co., Inc.), and images were acquired using a BioRad Molecular Imager Gel Doc XR system. FCS measurements were performed using an Alba FCS instrument at the Beckman Institute at the University of Illinois.