Articles

https://doi.org/10.1038/s41566-018-0324-z

Super-resolution enhancement by quantum image

scanning microscopy
Ron Tenne1,4, Uri Rossman1,4, Batel Rephael   1,4, Yonatan Israel1,2, Alexander Krupinski-Ptaszek3,
Radek Lapkiewicz3, Yaron Silberberg1 and Dan Oron   1*

The principles of quantum optics have yielded a plethora of ideas to surpass the classical limitations of sensitivity and resolu-
tion in optical microscopy. While some ideas have been applied in proof-of-principle experiments, imaging a biological sample
has remained challenging, mainly due to the inherently weak signal measured and the fragility of quantum states of light. In
principle, however, these quantum protocols can add new information without sacrificing the classical information and can
therefore enhance the capabilities of existing super-resolution techniques. Image scanning microscopy, a recent addition to
the family of super-resolution methods, generates a robust resolution enhancement without reducing the signal level. Here, we
introduce quantum image scanning microscopy: combining image scanning microscopy with the measurement of quantum pho-
ton correlation allows increasing the resolution of image scanning microscopy up to twofold, four times beyond the diffraction
limit. We introduce the Q-ISM principle and obtain super-resolved optical images of a biological sample stained with fluores-
cent quantum dots using photon antibunching, a quantum effect, as a resolution-enhancing contrast mechanism.

T
he diffraction limit, as formulated by Abbe, sets the attain- ISM each pixel in a detector array acts as a small pinhole in a confo-
able resolution in far-field optical microscopy to about half of cal laser-scanning microscope (CLSM), recording an image with a
the visible wavelength1, hindering its applicability in life sci- resolution twice as fine as the diffraction limit16,19. Thus, one can
ence studies at very small scales. Over the past two decades, several construct a super-resolved image without reducing the collected sig-
super-resolution methods have successfully overcome the diffrac- nal level20,21. Some variants offer the same resolution improvement
tion limit, including emission depletion microscopy, localization within an all-optical set-up, reducing the requirement for computa-
microscopy and structured illumination microscopy2–6. The con- tional power and fast imagers22–24. Although modest, this resolution
tinuous and rapid improvement in detector technology has enabled improvement is robust to changes in the sample and label type and
two more recent developments in the field of super-resolution can be integrated with additional microscopy modalities25,26.
microscopy, which are the centre of this work: quantum super-res- We present here a new super-resolution scheme, quantum image
olution microscopy and image scanning microscopy (ISM). As for scanning microscopy (Q-ISM), utilizing the measurement of quan-
the first, a surge of interest in super-resolution imaging based on tum correlations in an ISM architecture. During a confocal scan,
quantum optics concepts7–13, inspired and facilitated by the progress each pair of detectors in a detector array generates a sharp image
in high temporal resolution imagers, resulted in a few successful using photon correlations. These images are merged together to
proof-of-principle demonstrations7,8,14. The second, ISM, relies on surpass the resolution of both ISM and photon correlation measure-
a small array of fast detectors and offers a twofold enhancement in ments separately. By violating both the classical light and uniform
resolution15,16. Because ISM is compatible with a standard confocal illumination assumptions, we were able to obtain super-resolved
microscope architecture it has already been integrated into com- images of a biological sample based on a quantum optical effect,
mercial products. namely photon antibunching.
While all super-resolution modalities violate at least one of the
basic assumptions of Abbe’s theory, many rely on breaking more Resolution improvement in Q-ISM
than one. For instance, stimulated emission depletion and saturated Similarly to ISM, Q-ISM relies on a small modification of the stan-
structured illumination microscopy breach both the assumption of dard confocal microscope set-up (Fig. 1a). We use a standard con-
a linear response of a fluorophore to the excitation light and that of focal excitation scheme in which a pulsed blue laser beam (473 nm)
a uniform illumination field17,18. In contrast, the few demonstrations is focused through a high-numerical-aperture (NA) objective lens
of quantum super-resolution microscopy7,8,14 relied solely on violat- while the sample is scanned with a piezo stage. In a modified detec-
ing the implicit assumption, underlying Abbe’s derivation, that light tion scheme the sample fluorescence is collected through the same
behaves as waves rather than particles. ISM, as well, depends on objective, further magnified, and then imaged onto a honeycomb
violating a single assumption, uniform illumination of the sample. lattice fibre bundle (inset of Fig. 1a). Each fibre (14 in total) routes
An alternative approach to take advantage of the quantum prop- the light impinging on it to an individual singe-photon avalanche
erties of light to break Abbe’s limit is to enhance an already estab- detector (SPAD) (SPCM-AQ4C, Perkin-Elmer) feeding a time-
lished super-resolution method with the extra information held in correlated single-photon counting (TCSPC) card logging the arrival
a photon correlation measurement12. ISM is a natural candidate for times of the detected photons. Further information on the optical
this purpose as it does not compromise the collected signal level. In set-up is provided in the Methods.

Department of Physics of Complex Systems, Weizmann Institute of Science, Rehovot, Israel. 2Department of Physics, Stanford University, Stanford, CA,
1

USA. 3Institute of Experimental Physics, Faculty of Physics, University of Warsaw, Warsaw, Poland. 4These authors contributed equally: Ron Tenne, Uri
Rossman, Batel Rephael. *e-mail: dan.oron@weizmann.ac.il

116 Nature Photonics | VOL 13 | FEBRUARY 2019 | 116–122 | www.nature.com/naturephotonics

NaTure PHoTonIcS Articles
a b x1 Dettection where x is the scan coordinate. The product of the two terms in
x
Obj Exccitation equation (2) is depicted in Fig. 1b; a multiplication of the excitation
xi
ISM
M probability distribution (blue circle) and the detection probabil-
Dettector ity distribution of each detector (red circles) results in a narrower
DM
effective PSF (green circles) centred around 1 xa. The images from
2
the K different detectors can be shifted and summed into a single

TCSPC SPAD
T1
(1) K (1)
( 1
)
image G ISM (x ) = ∑ α=1 Gα x − xa . As an illustrative example,
2
Supplementary Section 2 presents a derivation of an explicit expres-
L1 x3 x2 sion of the ISM image in the case of a Gaussian PSF.
(1)
FB Although G ISM (x ) contains spatial frequencies two times larger
ISM
than the maximal spatial frequency contained in a widefield image,
c x1 ISM d 3,000 its resolution is enhanced only by a factor of ~ 2 . To extend the
Q-ISM image resolution to its Fourier limit one needs to perform Fourier
2,500 g (2)(0) = 0.95
0 95
Detector reweighting16. By digitally amplifying the high spatial frequency
2,000 content, one obtains an image with an up to twofold resolution
Photon pairs

1,500
enhancement (Supplementary Section 3).
g (2)(0) = 0.90 Light emitted from fluorophores, such as dye molecules, quan-
1,000 tum dots and solid-state defects, exhibits photon antibunching29–33,
500 a well-known effect in quantum optics. Each such fluorophore
x3 x3 g (2)(0) = 0.85 emits at most a single photon for every excitation pulse that is much
0 shorter than the fluorescence lifetime. Therefore, the number of
5
–500 0 500
Q-ISM
τ (ns) simultaneously detected photon pairs is lower than the number of
photon pairs at a delay of one or more pulses. The spatial distri-
Fig. 1 | Q-ISM principle of operation. a, The optical set-up used in this bution of missing events of simultaneously detected photon pairs
work. The pinhole in a standard confocal set-up is replaced with a fibre measured by a pair of detectors positioned at xa and xβ yields an
bundle (FB, shown in the inset) routing light into 14 individual SPADs. image of the photon correlation contrast following (Supplementary
Obj, objective lens; DM, dichroic mirror; T1, variable telescope; L1, lens. Section 1)
b, Schematic of the ISM method. The ISM PSF for each detector (green
M
circles) is a product of the excitation laser beam profile (blue circle) and
the detection probability distribution (red circles) centred on the detector’s
(2)
ΔGαβ (x ) ∝ ∑ [h (xi − x )] 2 × h (xa − (xi − x ))
i=1
(3)
position (black circles). For simplicity, the PSFs are drawn as cylindrically
symmetric, although the principle of operation of ISM and Q-ISM does not ×h (xβ − (xi − x ))
require this symmetry. c, Schematic of Q-ISM. The effective PSF for each
detector pair (orange circles) is a product of the two ISM PSFs of the two The effective two-photon absorption and two-photon detection
detectors (green circles). d, The unnormalized second-order correlation contrast results in a PSF, schematically illustrated in Fig. 1c (orange
function, G(2)(τ), generated for three positions in the scan shown in Fig. 3. circles), which is a product of two effective ISM PSFs centred around
1 1
Black dashed lines indicate the average value of incident photon pairs with x and 2 xβ (green circles). The photon correlation image is a sum
2 a
a time delay longer than one pulse. The calculated normalized second- of effective PSFs that are approximately centred at xi + 1 (xa + xβ )
4
order correlation function, g(2)(τ =​ 0), is shown next to each of the curves. and are narrower by a factor of two compared to the original PSF.
Similar to ISM, the K (K − 1) images of photon correlations can be
2
shifted and summed to obtain the merged Q-ISM image:
In the following, we outline the theoretical concept of super-
resolution in Q-ISM (described in full in Supplementary Section 1). K K
1
(2)  
To explore the resolution gain in Q-ISM in this type of set-up, let us −ISM (x ) = ∑
GQ(2) ∑ ΔGαβ x − (xa + xβ )  (4)
 4 
consider the case of M identical emitters positioned at xi (i =​  1,..., M) α = 1 β=α+ 1
and imaged by a unity magnification imaging system whose inco-
herent point spread function (PSF) is h(x). Applying a uniform illu-
mination field yields an intensity image As in the case of ISM, the resolution enhancement of GQ(2) −ISM (x )
(approximately two) can be extended further by performing Fourier
M reweighting. This can increase the resolution of a high signal-to-
G (1) (x ) ∝ ∑ h (x − xi ) (1) noise ratio (SNR) Q-ISM image up to four times beyond the diffrac-
i=1 tion limit (Supplementary Section 3). A more concrete mathematical
treatment of Q-ISM image formation from photon correlations for
Here we follow the notations used in ref. 11 indicating that the fluo- the case of a Gaussian PSF is provided in Supplementary Section 2.
rescence intensity can be viewed as the equal-time first-order Glauber’s To clarify the source of contrast in Q-ISM, Fig. 1d presents
correlation function27,28. In standard ISM, presented schematically unnormalized second-order correlation curves for three positions
in Fig. 1b, the sample is illuminated with a laser beam focused to a in a sample scan. The curves are constructed by performing pixel
(diffraction-limited) spot centred around x =​ 0 (blue circle)21. As the reassignment for the second-order correlation measurement for
sample position is scanned, the photoluminescence signal is recorded each detector pair, similar to the one performed in equation (4).
by an array of point-like detectors (filled black circles), whose image Each of the curves exhibits a plateau whose value is defined as
plane positions are denoted by xa (α =​1, ..., K). Neglecting the fluores- G (2) (∞) , marked by a dashed line, and a small, yet clearly visible,
cence Stokes shift, a single detector α obtains an intensity image antibunching dip at τ =​ 0. The value of a Q-ISM image at each posi-
tion is simply the difference ΔG (2) = G (2) (∞) − G (2) (0). Note that
M while the relative size of this dip (in proportion to the value of the
Gα(1) (x ) ∝ ∑ h (xi − x ) × h (xa − (xi − x )) (2) plateau) approaches zero for a large number of emitters, the abso-
i=1 lute magnitude of ΔG (2) increases with the number of emitters and

Nature Photonics | VOL 13 | FEBRUARY 2019 | 116–122 | www.nature.com/naturephotonics 117

Articles NaTure PHoTonIcS

a ×104 Fourier reweighted (FR) ISM, Q-ISM and FR Q-ISM analyses from
CLSM
a scan of a few quantum dots. The standard CLSM image (Fig. 2a)
3
was obtained by summing the intensities of all detectors, effec-
tively treating the fibre bundle as a semi-open pinhole in a confo-
2 cal microscope with an effective diameter of 0.6 Airy units (AU,
λ
1AU = 1.22 NA ). Here we observe a single unresolved elongated
1 patch containing fluctuations, probably resulting from quantum dot
blinking. By shifting the image obtained from each of the detectors,
a process termed pixel reassignment19 (Supplementary Section 6),
b ISM c FR ISM ×104
5 and summing those together, we generate an ISM image (Fig. 2b). In
addition to the enhanced resolution, temporal fluctuations in fluo-
4
rescence during the sample scan (Fig. 2a) are smoothed, becoming
3 unnoticeable in ISM (Fig. 2b), due to the delay accrued between the
2 shifted detector images. The FR ISM image, calculated by applying
a filter on the ISM image shown in Fig. 2b in the Fourier domain, is
1
shown in Fig. 2c (Supplementary Section 3).
To generate a Q-ISM image we analysed the same data set to
d e
Q-ISM FR Q-ISM
100 calculate the second-order photon correlation (Fig. 1d). These
correlations were analysed for each position along the scan using
the photon arrival times in every pair of detectors (Supplementary
Section 5). As described above, taking the difference between the
50
coincident and delayed photon pair detections yields an antibunch-
(2)
ing contrast image for each detector pair, ΔGαβ (x ) , which can be
merged according to equation (4) into a Q-ISM image, GQ(2) −ISM (x ) ,
0 shown in Fig. 2d (Supplementary Sections 5 and 6). Finally, FR
f was performed on the Q-ISM image (Fig. 2e) to flatten the spa-
tial frequency response of the imaging technique and bring the
resolution closer to the theoretical, fourfold enhancement limit
Photons (×104)

Photon pairs

4 100
(Supplementary Section 3). For a clearer comparison of the resolu-
tion improvement, Fig. 2f presents interpolated cross-sections from
2 50
the four images in Fig. 2b–e.
To perform a quantitative assessment of the resolution improve-
0
0.2 0.4 0.6 0.8
0 ment we imaged a few tens of isolated fluorescent beads and derived
Position (μm) the PSFs for wide-field, ISM, FR ISM, Q-ISM and FR Q-ISM imag-
ing. From the width of the PSF we extracted the expected resolu-
Fig. 2 | Resolving emitters with Q-ISM. A 1.5 μ​m ×​ 1.5 μ​m confocal scan tion according to the Rayleigh criterion (Supplementary Section 7)
(50 nm steps, 200 ms pixel dwell time) of a sample of a cluster of CdSe/ as 454 nm in the standard wide-field microscope, 352 nm in ISM,
CdS/ZnS quantum dots. a, The summed intensity over the whole detector 231 nm in FR ISM, 261 nm in Q-ISM and finally 194 nm in the FR
array per scan position. b, ISM image, GISM(1)
(x) : the scan image from Q-ISM images. The resolution enhancement factors compared to
each detector is shifted before summation. c, A Fourier reweighted ISM the wide-field image are 1.29, 1.97, 1.74 and 2.34 for the ISM, FR
image obtained from b (Supplementary Section 3). The colour bars for ISM, Q-ISM and FR Q-ISM images respectively. The fact that both
a–c represent the number of detected photon counts. d, Q-ISM image, the ISM and Q-ISM images do not fulfil the theoretical enhance-
GQ(2)
−ISM (x ) : the antibunching signal image of each detector pair is shifted
ment limit of 2 and 2, respectively, is probably due to the non-
and summed (equation (4) and Supplementary Sections 1 and 5). e, A negligible detector size in our system (a de-magnified diameter
Fourier reweighted (FR) image obtained from d (Supplementary Section 3). of ~0.25λ). In FR ISM and especially in FR Q-ISM, the resolution
The colour bars for d and e represent the number of missing detected enhancement is further limited by the finite SNR of the unfiltered
photon pairs. For clarity, the maximal values of the Fourier reweighted images. Although, according to theoretical considerations, the reso-
images in c and e are scaled to that of the ISM (b) and Q-ISM (d) images, lutions of FR ISM and Q-ISM are equal, in the current experimental
respectively. f, Interpolated cross-sections for the dotted white line shown realization FR ISM reaches a somewhat higher resolution. However,
in e for four different analyses: ISM (dashed blue), FR ISM (solid blue), comparing Fig. 2c,d one can see that a significant portion of the
Q-ISM (dashed red) and FR Q-ISM (solid red). Scale bar, 0.5 μ​m. light intensity in the PSF of FR ISM is distributed in secondary rings
around the emitter. These rings are generated due to the frequency
cutoff enforced by Fourier reweighting (Supplementary Section 3).
While the intensity of these side rings does not reduce the resolu-
its visibility remains unchanged (see Supplementary Section 4 for tion according to the Rayleigh criterion, it can degrade the visibility
detailed discussion of SNR in photon correlation measurements).
(2)
of features in a biological scene, which is usually quite dense.
The normalized second-order correlation dip, g (2) (0) = G(2) (0) , is
G (∞)
shown next to each curve. Quantum microscopy of a fixed cell sample
We further demonstrate the Q-ISM method by imaging a biologi-
Demonstration of resolution enhancement in Q-ISM cal sample of microtubules labelled with quantum dots in fixed
To demonstrate the concept of Q-ISM we used the microscope set- 3T3 cells (see Methods). Figure 3 presents a super-resolved FR
up described in the previous section to image a sample of quantum Q-ISM image of a 3 μ​m  ×​  3  μ​m area as well as CLSM, ISM, FR ISM
dots randomly dispersed on a substrate. The detection times of pho- and Q-ISM images analysed from the same data set. Although
tons collected during the scan were parsed and analysed to gener- the SNR is lower in the Q-ISM and FR Q-ISM images (Fig. 3d,e),
ate the ISM and Q-ISM images (see Methods and Supplementary all the visible features in the CLSM (Fig. 3a) and ISM (Fig. 3b)
Section 5 for further details). Figure 2 compares the CLSM, ISM, images are present with a finer resolution. This observation

118 Nature Photonics | VOL 13 | FEBRUARY 2019 | 116–122 | www.nature.com/naturephotonics

NaTure PHoTonIcS Articles
a CLSM b ISM c FR ISM
×104 ×104 ×104
2 4 6 8 2 4 6 8 2 4 6

d Q-ISM e FR Q-ISM
0 50 100 150 0 50 100

Fig. 3 | Q-ISM of labelled microtubule cell samples. a–e, Images analysed from a confocal scan (50 nm steps, 100 ms pixel dwell time) of a 3 μ​m ×​ 3 μ​m
section of microtubules in a fixed 3T3 cell labelled with fluorescent quantum dots (QDot 625, Thermo Fisher): CLSM image (a), ISM image (b), FR ISM
image (c), Q-ISM image (d) and FR Q-ISM image (e). The colour bars in a–e follow the same scheme as in Fig. 2a–e, respectively. For clarity, the maximum
intensity of the Fourier reweighted images shown in c and e are scaled to the maximum intensity of their source images in b and d, respectively. Enlarged
images of a section in each image, framed by the white dashed line, are shown below each image. Scale bar, 0.5 μ​m.

demonstrates the valuable information contained in photon corre- same sample shown in Fig. 3, for different imaging planes, shifted
lation data even though it is based on rare events. Supplementary by 0.4 μ​m relative to each other. Defocusing the objective results
Section 4 provides a more elaborate discussion regarding SNR in in widening of both the excitation and detection PSFs. Therefore
Q-ISM. As expected from theoretical considerations, the FR ISM a decrease in the integrated signal level accompanies a blurring of
image (Fig. 3c) provides a similar transverse resolution to that the image. Figure 4a–d show the ISM images obtained for objec-
of the unfiltered Q-ISM image (Fig. 3d). However, in the rather tive positions at −​0.8, −​0.4, 0 and 0.4 μ​m, respectively (0 designates
dense scene shown in these images, the addition of a pronounced the position of the collection optimal focus). As expected from a
ringing artefact by Fourier reweighting can obscure faint fea- confocal microscope the fine image features are blurred and their
tures adjacent to bright ones. Another advantage of Q-ISM over total intensities decrease as we move from the focus position19,20.
CLSM, ISM and FR ISM, presented in the following section, is an In comparison, the contrast of Q-ISM images, presented in Fig. 4e–
improved axial resolution. h, has a stronger dependence on defocusing, rejecting more of the
out-of-focus contributions.
Z-sectioning in Q-ISM Heuristically, the sharper decrease in Q-ISM image intensity
Q-ISM offers an improved resolution compared to ISM and FR occurs because the contrast results from missing events in which
ISM, not only in the transverse directions but also in the axial two excitation photons are absorbed and two emitted photons are
direction. A theoretical discussion and numerical results compar- detected in a small detector array. The probability of those inde-
ing the z-sectioning capabilities of Q-ISM and ISM are provided in pendent events occurring simultaneously decreases substantially
Supplementary Section 8. To experimentally explore the axial reso- with defocusing because both the excitation power density and the
lution enhancement, Fig. 4 presents images of a thin area from the probability to detect over the small array decrease (Supplementary

Nature Photonics | VOL 13 | FEBRUARY 2019 | 116–122 | www.nature.com/naturephotonics 119

Articles NaTure PHoTonIcS

a Z = –0.8 µm b Z = –0.4 µm c Z = 0 µm d Z = 0.4 µm ×104

ISM
4

e f g h
200

150
Q-ISM

100

50

i ×106

3
6,000

Photon pairs
Photons

2 4,000

1 2,000

0
0
–1 –0.5 0 0.5 1
z (µm)

Fig. 4 | Resolving power of Q-ISM in the axial dimension. a–d, ISM images of a 1 μ​m ×​ 1 μ​m (50 nm pixel size) axially thin region in the 3T3 cells sample
at objective defocus positions of −​0.8, −​0.4, 0 and 0.4 μ​m, respectively. 0 is the sample focus plane. The colour bar indicates the number of detected
photons. e–h, Q-ISM images analysed from the same measurement as in a–d, respectively. The colour bar indicates the number of missing photon pairs
per scan position. i, Integrated signal of the ISM and Q-ISM images over the area enclosed by the green dashed rectangle in a, shown as blue and red
circles respectively. Violet vertical lines show the defocus positions of the images shown in a–h. Scale bars, 0.25 μ​m.

Section 8). This effect is highlighted in Fig. 4i, which displays the Relying on the photon correlation contrast poses some additional
intensity sum for ISM (blue) and Q-ISM (red) over a thin section in constraint on sample labelling. First, an isolated fluorescent probe
the sample versus the defocusing position. Although the improved should present some degree of antibunching. Notably, however, this
axial resolution is clearly visible in Fig. 4, the non-negligible sam- is not a particularly limiting condition, because commercial dye
ple thickness prevents a quantitative estimation of the resolution molecules, quantum dots, solid-state defects and some fluorescent
enhancement. To overcome this issue we analysed images of a thin proteins fulfil this requirement33,40,41. Since the photon correlation
sample of quantum dots spin-coated on a coverslip obtained under SNR increases monotonically with fluorescence quantum yield and
similar experimental conditions (Supplementary Section 8). The excitation repetition rate, labels should exhibit a high quantum yield
full-width at half-maximum (FWHM) of the integrated intensity under close-to-saturation conditions. Photostability for a duration
from a wide area in the case of Q-ISM (1 μ​m) is narrower by a factor on the scale of a second, necessary for high SNR and to avoid scan-
of ~2.1 than the FWHM for ISM (2.1 μ​m). ning artefacts, has been demonstrated for colloidal quantum dots,
solid-state defects and a variety of dye molecules routinely used in
Discussion bio-imaging33,39,41. Finally, to achieve high photon rates under satu-
Q-ISM is compatible with standard confocal microscopy, with a ration conditions, the emitter’s excited-state lifetime should prefer-
few additional requirements from the experimental set-up and ably be below a microsecond.
the fluorescent markers. First, the set-up requires a small array Unlike some of the previous implementations of imaging
of detectors with nanosecond-scale temporal resolution, such assisted by a photon correlation measurement12, Q-ISM does not
as a small monolithic SPAD array or a fibre bundle detector34–36. rely on spatial sparsity of emitters. Because each emitter adds to the
The latter set-up avoids inter-detector crosstalk, a crucial advan- occurrence of missing photon pairs, dense scenes, such as those
tage for accurate measurement of antibunching. To speed up the presented in Fig. 3, can be imaged. However, increasing the label-
acquisition process, conceptually similar ideas11 may be imple- ling density of the sample does not improve the SNR because both
mented in a wide-field configuration by employing a large-area the number of missing photon pairs and its standard error grow
SPAD array37. Alternatively, using a large SPAD array one can linearly with emitter density (Supplementary Section 4). In fact,
also perform a multifocal adaptation38 of the Q-ISM concept. As given a magnification and pixel size, the saturation of the single-
for the excitation source, either a pulsed or a continuous wave photon detectors sets an upper limit on the sample density, and
(c.w.) laser can be used without affecting the method’s resolution crossing that limit can degrade the image. In principle, the Q-ISM
or SNR. Here, we chose to work with a pulsed excitation source method could be extended to higher correlation orders to achieve
because it is favourable in avoiding photobleaching under saturat- an even higher resolution, but the SNR of correlation orders higher
ing illumination39. than two will decrease with labelling density (Supplementary

120 Nature Photonics | VOL 13 | FEBRUARY 2019 | 116–122 | www.nature.com/naturephotonics

NaTure PHoTonIcS Articles
Section 7). Therefore, applying those as the contrast of ISM may References
prove challenging. 1. Abbe, E. Beiträge zur Theorie des Mikroskops und der mikroskopischen
The signal of Q-ISM relies on missing two-photon events: the Wahrnehmung. Arch. für Mikroskopische Anat. 9, 413–418 (1873).
2. Hell, S. W. & Wichmann, J. Breaking the diffraction resolution limit by
absence of a detected photon pair emission from a single fluoro- stimulated emission: stimulated-emission-depletion fluorescence microscopy.
phore that absorbs two photons at the same pulse (Supplementary Opt. Lett. 19, 780–782 (1994).
Section 1). Since in far-field microscopy the probability of single- 3. Gustafsson, M. G. L. Surpassing the lateral resolution limit by a factor of two
photon detection per fluorophore and per pulse is much less than using structured illumination microscopy. J. Microsc. 198, 82–87 (2000).
4. Rust, M. J., Bates, M. & Zhuang, X. Sub-diffraction-limit imaging by
unity, missing events of two-photon detection are rather rare. As a
stochastic optical reconstruction microscopy (STORM). Nat. Methods 3,
result, the acquisition time required for a high-SNR Q-ISM is appre- 793–795 (2006).
ciably longer than that of an ISM image (Supplementary Section 4). 5. Betzig, E. et al. Imaging intracellular fluorescent proteins at nanometer
However, it is worth noting that the measurement of photon corre- resolution. Science 313, 1642–1645 (2006).
lations is complementary to the standard ISM data acquired simul- 6. Dertinger, T., Colyer, R., Iyer, G., Weiss, S. & Enderlein, J. Fast, background-
free, 3D super-resolution optical fluctuation imaging (SOFI). Proc. Natl Acad.
taneously. Thus, synthesis of the two data sets, using both the ISM Sci. USA 106, 22287–22292 (2009).
data and the Q-ISM data as constraints, may be a valuable path to 7. Schwartz, O. et al. Superresolution microscopy with quantum emitters. Nano
enhance resolution in ISM. Under the current experimental condi- Lett. 13, 5832–5836 (2013).
tions, a 1 ms pixel dwell time is sufficient to achieve an SNR above 8. Gatto Monticone, D. et al. Beating the Abbe diffraction limit in confocal
unity in the measurement of photon correlations (Supplementary microscopy via nonclassical photon statistics. Phys. Rev. Lett. 113, 143602 (2014).
9. Tsang, M. Quantum limits to optical point-source localization. Optica 2,
Section 4). While this signal level, in itself, is not enough to con- 646–653 (2015).
struct a super-resolved image such as those shown in Figs. 2 and 3, 10. Tsang, M., Nair, R. & Lu, X.-M. Quantum theory of superresolution for two
it contains valuable information that is not accessible in standard incoherent optical point sources. Phys. Rev. X 6, 031033 (2016).
ISM imaging. 11. Classen, A., von Zanthier, J., Scully, M. O. & Agarwal, G. S. Superresolution
A few variants of ISM have already extended the lateral resolu- via structured illumination quantum correlation microscopy. Optica 4,
580 (2017).
tion or the z-sectioning capabilities of ISM by taking advantage 12. Israel, Y., Tenne, R., Oron, D. & Silberberg, Y. Quantum correlation enhanced
of the sample’s nonlinear response. Gregor and colleagues applied super-resolution localization microscopy enabled by a fibre bundle camera.
a two-photon excitation scheme to induce z-sectioning and Nat. Commun. 8, 14786 (2017).
gain a high penetration depth in an all-optical variant of ISM26, 13. Aßmann, M. Quantum-optically enhanced STORM (QUEST) for multi-
whereas Laporte and colleagues harnessed the emitters’ saturation emitter localization. Sci. Rep. 8, 7829 (2018).
14. Cui, J.-M., Sun, F.-W., Chen, X.-D., Gong, Z.-J. & Guo, G.-C. Quantum
to achieve higher lateral resolution25. Note that in both of these statistical imaging of particles without restriction of the diffraction limit.
examples the excitation is nonlinear, while the detection pro- Phys. Rev. Lett. 110, 153901 (2013).
cess remains linear. In contrast, Q-ISM relies on missing photon 15. Sheppard, C. J. R. Superresolution in confocal imaging. Optik 80,
pair events: events of absorption and detection of two photons. 53–54 (1988).
Therefore, the signal measured in Q-ISM depends not only on the 16. Müller, C. B. & Enderlein, J. Image scanning microscopy. Phys. Rev. Lett. 104,
198101 (2010).
excitation PSF squared, but also on the second power of the detec- 17. Rego, E. H. et al. Nonlinear structured-illumination microscopy with a
tion PSF (equation (3)). The additional nonlinearity in detection photoswitchable protein reveals cellular structures at 50-nm resolution.
translates into higher available spatial frequencies and therefore Proc. Natl Acad. Sci. USA 109, E135–E143 (2012).
a possibility for higher resolution, up to a fourfold improvement 18. Blom, H. & Widengren, J. Stimulated emission depletion microscopy.
Chem. Rev. 117, 7377–7427 (2017).
over the diffraction limit (Supplementary Section 3). In that
19. Sheppard, C. J. R., Mehta, S. B. & Heintzmann, R. Superresolution by image
sense, Q-ISM realizes a theoretical concept, introduced by Hell scanning microscopy using pixel reassignment. Opt. Lett. 38, 2889 (2013).
and colleagues over two decades ago, of detecting only photon 20. Gu, M. & Sheppard, C. J. R. Confocal fluorescent microscopy with a
pairs with both photons emanating from the same position in a finite-sized circular detector. J. Opt. Soc. Am. A. 9, 151 (1992).
confocal set-up42,43. 21. Ward, E. N. & Pal, R. Image scanning microscopy: an overview. J. Microsc.
266, 221–228 (2017).
22. York, A. G. et al. Instant super-resolution imaging in live cells and embryos
Conclusions via analog image processing. Nat. Methods 10, 1122–1126 (2013).
In summary, Q-ISM combines the super-resolving capabilities of 23. Roth, S., Sheppard, C. J., Wicker, K. & Heintzmann, R. Optical photon
both quantum correlation measurements and ISM to achieve an reassignment microscopy (OPRA). Opt. Nanoscopy 2, 5 (2013).
up to fourfold enhanced resolution, significantly more than both 24. De Luca, G. M. R. et al. Re-scan confocal microscopy: scanning twice for
better resolution. Biomed. Opt. Express 4, 2644–2656 (2013).
methods separately. This combination enabled imaging of a bio- 25. Laporte, G. P. J., Stasio, N., Sheppard, C. J. R. & Psaltis, D. Resolution
logical sample, fixed cells labelled with single-photon emitters, enhancement in nonlinear scanning microscopy through post-detection
relying only on the antibunching contrast. Because Q-ISM pro- digital computation. Optica 1, 455 (2014).
vides information complementary to the conventional ISM data 26. Gregor, I. et al. Rapid nonlinear image scanning microscopy. Nat. Methods
measured simultaneously, we believe that it may offer a substan- 14, 1087–1089 (2017).
27. Glauber, R. J. The quantum theory of optical coherence. Phys. Rev. 130,
tial and practical improvement to ISM in terms of resolution and 2529–2539 (1963).
background. An algorithmic combination of the high resolution 28. Gerry, C. C. & Knight, P. L. Introductory Quantum Optics (Cambridge
contained in a Q-ISM image with the high SNR provided by a stan- University Press, Cambridge, 2005).
dard ISM image may lead to an image significantly surpassing each 29. Kimble, H. J., Dagenais, M. & Mandel, L. Photon antibunching in resonance
technique separately. fluorescence. Phys. Rev. Lett. 39, 691–695 (1977).
30. Basche’, T., Moerner, W. E., Orrit, M. & Talon, H. Photon antihunching in the
fluorescence of a single dye molecule trapped in a solid. Phys. Rev. Lett. 69,
Online content 1516–1519 (1992).
Any methods, additional references, Nature Research reporting 31. Michler, P. et al. A quantum dot single-photon turnstile device. Science 290,
summaries, source data, statements of data availability and asso- 2282–2285 (2000).
ciated accession codes are available at https://doi.org/10.1038/ 32. Brouri, R., Beveratos, A., Poizat, J.-P. & Grangier, P. Photon antibunching in
the fluorescence of individual color centers in diamond. Opt. Lett. 25,
s41566-018-0324-z. 1294–1296 (2000).
33. Grußmayer, K. S. & Herten, D.-P. Time-resolved molecule counting by
Received: 19 June 2018; Accepted: 15 November 2018; photon statistics across the visible spectrum. Phys. Chem. Chem. Phys. 19,
Published online: 17 December 2018 8962–8969 (2017).

Nature Photonics | VOL 13 | FEBRUARY 2019 | 116–122 | www.nature.com/naturephotonics 121

Articles NaTure PHoTonIcS
34. Roider, C., Ritsch-Marte, M. & Jesacher, A. High-resolution confocal Raman consolidator grant ColloQuanto, ERC grant QUAMI, the ERC-POC project ‘SFICAM’,
microscopy using pixel reassignment. Opt. Lett. 41, 3825–3828 (2016). the ICore program of the ISF, the Crown Photonics Center and the Israeli ministry
35. Portaluppi, D., Conca, E. & Villa, F. 32 x 32 CMOS SPAD imager for gated of science Tashtiyot program. R.L. and A.K.-P. acknowledge the hospitality of the
imaging, photon timing, and photon coincidence. IEEE J. Sel. Top. Quantum Weizmann Institute of Science and the support of National Science Centre (Poland)
Electron. 24, 3800706 (2018). grants nos. 2015/17/D/ST2/03471 and 2015/16/S/ST2/00424, the Polish Ministry
36. Castello, M. et al. Image scanning microscopy with single-photon detector of Science and Higher Education, and the Foundation for Polish Science under the
array. Preprint at https://doi.org/10.1101/335596 (2018). FIRST TEAM project ‘Spatiotemporal photon correlation measurements for quantum
37. Antolovic, I. M., Burri, S., Bruschini, C., Hoebe, R. A. & Charbon, E. SPAD metrology and super-resolution microscopy’ cofinanced by the European Union under
imagers for super resolution localization microscopy enable analysis of fast the European Regional Development Fund.
fluorophore blinking. Sci. Rep. 7, 44108 (2017).
38. York, A. G. et al. Resolution doubling in live, multicellular organisms
via multifocal structured illumination microscopy. Nat. Methods 9, Author contributions
R.T., Y.I., Y.S. and D.O. proposed and designed the experiment. R.T., U.R., B.R., Y.I. and
749–754 (2012).
R.L. performed the experimental work. R.T., U.R., B.R., A.K.-P. and R.L. performed the
39. Schwartz, O. et al. Colloidal quantum dots as saturable fluorophores.
data analysis. R.T. wrote the manuscript with significant contributions from all authors.
ACS Nano 6, 8778–8782 (2012).
40. Sýkora, J. et al. Exploring fluorescence antibunching in solution to determine
the stoichiometry of molecular complexes. Anal. Chem. 79, 4040–4049 (2007). Competing interests
41. Eisaman, M. D., Fan, J., Migdall, A. & Polyakov, S. V. Invited review article: The authors delcare no competing interests.
Single-photon sources and detectors. Rev. Sci. Instrum. 82, 071101 (2011).
42. Hänninen, P. E., Schrader, M., Soini, E. & Hell, S. W. Two‐photon excitation
fluorescence microscopy using a semiconductor laser. Bioimaging 3, Additional information
70–75 (1995). Supplementary information is available for this paper at https://doi.org/10.1038/
43. Hell, S. W., Soukka, J. & Hänninen, P. E. Two‐ and multiphoton detection as s41566-018-0324-z.
an imaging mode and means of increasing the resolution in far‐field light Reprints and permissions information is available at www.nature.com/reprints.
microscopy: a study based on photon‐optics. Bioimaging 3, 64–69 (1995).
Correspondence and requests for materials should be addressed to D.O.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in
Acknowledgements
published maps and institutional affiliations.
The authors thank Y. Ebenstein for the preparation of biological samples and S. Itzhakov
for synthesizing the quantum dots used in this work. This work was supported by ERC © The Author(s), under exclusive licence to Springer Nature Limited 2018

122 Nature Photonics | VOL 13 | FEBRUARY 2019 | 116–122 | www.nature.com/naturephotonics

NaTure PHoTonIcS Articles
Methods 3 μ​W, coupled to a single-mode fibre and focused through a 1.4 NA objective lens
Sample preparation. CdSe/CdS/ZnS quantum dots were prepared in a colloidal (Plan Apo Vc 100, Nikon) into a stationary PSF. A two-axis piezo stage (P-542.2SL
synthesis. Details regarding the synthesis are provided in the supplementary XY, Physik Instrumente) scanned the sample row-by-row as it was illuminated. The
information of ref. 7. The dimensions of the resulting rod-shaped quantum dots are fluorescence was collected by the same objective lens and filtered with dichroic
~6 ×​  6  ×​ 10 nm (Supplementary Section 9). The fluorescence was centred around mirrors and filters (FF509-FDi01, SP01-785RS, BLP01-532R, Semrock). A Galilean
617 nm and presented a 0.62 quantum yield for a 530 nm excitation wavelength. beam expander (BE05-10-A, Thorlabs) was placed after the relay lens to magnify
The quantum dots were diluted in a 3 wt% solution of poly(methylmethacrylate) in the imaged fluorescence spot onto a fibre bundle (Art Photonics). This fibre
toluene and spin-coated onto a microscope coverslip. bundle consisted of multimode 100/110 μ​m core/clad fibres, fused together on the
Fixed cell samples were prepared as follows. NIH 3T3 cells were grown to 80% entrance side (inset of Fig. 1a). On the exit side the fibres were arranged to fan out
confluency and fixed by 15 min incubation in cytoskeleton buffer (CB) (10 mM to individual multimode fibres; 14 of those guided light from the image plane to
Mes (pH 6.2), 140 mM NaCl, 2.5 mM EGTA, 5 mM MgCl2) containing 11% individual single-photon avalanche photodiodes (SPCM-AQ4C, Perkin-Elemer).
sucrose, 3.7% paraformaldehyde, 0.5% glutaraldehyde and 0.25% Triton. Fixation For a detailed characterization of the fibre bundle set-up see the supplementary
was stopped with 0.5 mg ml–1 sodium borohydride in CB for 8 min, followed by information in ref. 12.
washing with PBS and a 1 h blocking step with 2% BSA in PBS. Fixed cells were
incubated with a 1:500 dilution of DM1A anti-α​-tubulin monoclonal antibody Data acquisition and analysis. A time-correlated single-photon counting board
(Sigma) in PBS with 2% BSA and washed three times in PBS. QD625-labelled goat was used for data acquisition in absolute timing mode (DPC-230, Becker & Hickl).
F(ab)2, anti-mouse IgG antibodies (HL) (Invitrogen) were diluted 1:400 in PBS An excitation pulse trigger was synchronized and recorded at every 40th pulse
with 6% BSA and applied to cells for 1 h. Cells were then dehydrated by sequential (0.5 MHz). The correlation analysis, ISM, Q-ISM and Fourier reweighting were
washing for several seconds in 30, 70, 90 and 100% ethanol. Finally, cells were spin- implemented in a MATLAB script, post-processing the acquired data. Further
coated (500 r.p.m.) with 1 mg ml–1 polyvinyl alcohol (PVA). details about analysis are provided in Supplementary Sections 3, 5 and 6.

Microscope set-up. A custom-built set-up around a commercial optical Data availability

microscope (Zeiss Axiovert 135) was used to image fluorescent samples in confocal The raw data that support the findings of this study are available in the figshare
mode. The sample was illuminated with a 473 nm pulsed picosecond laser diode repository under the name ‘QISM_SoftwareAndData_zip’ and with the identifier
(EPL475, Edinburgh Instruments) with a 20 MHz repetition rate and a power of https://doi.org/10.6084/m9.figshare.7241294.v1.

Nature Photonics | www.nature.com/naturephotonics