A method for single molecule localization microscopy at Ångstrom resolution based on conventional fluorophore-labeled antibody probes

Abstract

Method for single molecule localization microscopy (SMLM) based on conventional fluorescent binder probes, including primary or secondary antibodies, that are applied in a gradual stepwise fashion to a fixed biological sample. In each step of the process, a low dose of the probe is applied in solution to the fixed biological sample to sparsely label target epitopes followed by sample washing to remove the free probe. The remaining bound fluorescent probes that are sparsely distributed over the sample are then imaged until they are photobleached. The high light collection efficiency from each probe in this approach (limited only by photobleaching) enables few Ångstrom lateral resolution and few nm axial resolution. Photobleaching of the probes at the end of each step generates a clean slate for application of the next step in the process, as well as the possibility for applying one or more probes in a consecutive fashion to the same sample for serial SMLM.

Description

A METHOD FOR SINGLE MOLECULE LOCALIZATION MICROSCOPY AT ANGSTROM RESOLUTION BASED ON CONVENTIONAL FLUOROPHORE- LABELED ANTIBODY PROBES

[0001] The invention is directed to a method for single molecule localization microscopy (SMLM) down to Angstrom-level resolution based on repeated steps of sparse labeling, washing, and imaging / photobleaching of a conventional binder, such as a primary or secondary antibody, conjugated with one or more standard fluorophores.BACKGROUND

[0002] Single Molecule Localization Microscopy (SMLM) is based on the sparse detection of fluorescent binder probes that label specific target epitopes within a fixed sample. Sparsity of detection is necessary in order to isolate and super-resolve individual fluorescent binder probes. There are three different strategies that have been reported for SMLM (all first published in 2006) that enable sparse detection on a widefield microscope setup: a. Photo-activation: PhotoActivated Localization Microscopy (PALM) i. Betzig et al., Science 313, 1642 (2006) b. Chemical quenching: STochastic Optical Reconstruction Microscopy (STORM) i. STORM: Rust, Bates, & Zhuang, Nature Methods 3, 793 (2006) ii. dSTORM: Heilemann et al., Angew Chem Int Ed Engl 47, 6172 (2008) c. Transient binding: Point Accumulation for Imaging in Nanoscale Topography (PAINT) i. PAINT: Sharonov and Hochstrasser, PNAS 103, 18911 (2006) ii. DNA-PAINT: Jungmann et al., Nature Methods 11, 313 (2014) iii. IRIS: Kiuchi et al., Nature Methods 12, 743 (2015) iv. Fluorogenic DNA-PAINT: Chung et al., Nature Methods 19, 554 (2022) v. RESI: Reinhardt et al., Nature 617, 711 (2023)

[0003] PALM is based on cycles of sparse activation of a subset of photo-activatable fluorophores in the sample followed by imaging of the fluorophores until they are all photobleached. Photo-activatable fluorophores are either genetically encoded fluorophores (paGFP, paCherry, mEos family, etc.) that are expressed in cells as a fusion to a protein target of interest, or are available as small molecule dyes that can be conjugated to an antibody (Grimm et al., Nat Meth 13, 985, 2016). PALM is therefore limited to the number of fluorescent channels available for the photo-activatable fluorophores, so at the very most roughly five different labels can be simultaneously detected in a single biological sample. One advantage of PALM is that all of the fluorescent signal from a single probe can be detected before its irreversible photobleaching, yielding the highest possible localization accuracy limited only by fluorophore photostability. However, both the genetically encoded photo- activatable fluorophores as well as the small molecule photo-activatable dyes do not generate a strong enough detection signal before they photobleach to reach better than roughly 10 - 20 nm resolution. If more than one photo-activatable fluorophore is used to label a target epitope, the localization accuracy obtained for each individual fluorophore is typically insufficient to confidently assign all photo-activation events to the exact same probe (assignment problem). For this reason, only a single fluorophore tag is typically used to label each protein target.

[0004] STORM is based on the stochastic blinking of fluorophores bound to primary or secondary antibodies that are highly quenched due to the use of a quenching buffer with specific redox potential. The most widely used variant of STORM is called dSTORM (Heilemann et al. 2008). While a few other dyes have been demonstrated for use with dSTORM, Alexa647 is the most optimal and most widely used. As for PALM, the number of different labels is limited by the number of distinct fluorescence channels, so at most roughly five different fluorophore probes can be simultaneously used on the same sample. A major limitation for STORM is the following assignment problem: Temporally displaced blinks from the same probe are not resolved well enough to reliably assign them all to the same probe, thereby limiting overall localization precision. The assignment problem and the fact that not all individual fluorophores exhibit the same number of blinks generates a statistical blinking noise or detection bias in the images (not present for PALM) and hampers the accuracy of assessing the total labeled epitopes across the field of view under consideration.

[0005] Serial staining with STORM (Klevanski et al., Nat Comm 11, 1552, 2020) or the combination of STORM and PALM (Paupiah et al., Biol Imaging 3:el4, 2023), can beachieved by iteratively staining the sample in a conventional manner (i.e. binding the antibody to the sample until near saturation of the epitope), washing the sample, and acquiring the STORM or PALM images until all labeled epitopes have been detected and photobleached. Here, specific fluorophores are further required that can be chemically quenched (STORM) or photo-activated (PALM) to enable the sparse detection of isolated probes required for SMLM acquisition at each step. Complete photobleaching of the probes at the end of each step then creates a blank slate for another round of conventional staining (binding to near saturation), washing, and imaging for a different probe labeling a different target epitope.

[0006] PAINT is based on the super-resolution of sparse, transient binding events of fluorophore-labeled probes that either directly bind to target epitopes in the sample (IRIS), or that bind to antibodies that label the target epitopes (DNA-PAINT). As for dSTORM, PAINT is also limited by a similar assignment problem. For PAINT, localization accuracy for an individual transient binding event is limited by the residence time of the event, with the achieved accuracy typically insufficient to confidently assign two or more binding events occurring in the same location to the same target epitope. As well, a binding noise or detection bias is present in the images generated simply from the statistics of the random binding to each individual epitope (e.g. some epitopes will be visited more often than others), leading to a “speckled” image reconstruction. This detection bias can be reduced through longer integration, but only when every binding site in the sample is visited many times. One advantage of PAINT -based approaches is their immediate applicability for serial superresolution, as the transient binders can be efficiently removed from the sample by washing, allowing application of multiple different probes on the same sample. For IRIS, only some tens of probes have so far been generated. For DNA-PAINT, probes have also only been designed for some tens of specificities. For DNA-PAINT, two probes are required for each specificity, an oligo-coupled antibody with a sequence unique for that antibody and its complementary oligo labeled with a fluorophore (imager strand). While IRIS probes can be applied and acquired in an apparently unlimited fashion to the sample, serial DNA-PAINT over many cycles of different probes requires careful optimization of the oligo sequences to avoid cross-talk of a given imager strand with the wrong oligo-conjugated antibody. The high light dose required for SMLM can also lead to cleavage of the oligo tags that label the antibodies for DNA-PAINT. For efficient SMLM acquisition with DNA-PAINT or IRIS, only probes with sufficiently high affinity (low binding constant, KD) can be employed, asotherwise the amount of probe required to apply to the sample creates a background signal that is too high to robustly detect the individual localizations. To reduce this background, optical sectioning techniques like total internal reflection fluorescence (URF) or HILO microscopy are typically employed, but even here there is also ultimately a limit to binders with sufficiently high affinity (low KD).

[0007] To remove the background of free probe in DNA-PAINT, an approach called fluorogenic DNA-PAINT has recently been developed based on quenching of the dye on the imager strand by a quenching molecule when the imager strand is not bound to its target complementary oligo (Chung et al., Nature Methods 19, 554, 2022). When bound to its target oligo, the quencher is moved to a large enough distance from the dye, allowing the latter to fluoresce. Here, quenching of the free imager strand is not perfect, but can nevertheless reduce the background fluorescence of the free imager strand by roughly an order of magnitude. One disadvantage of this strategy is that a longer sequence is required for the imager strand, making a complementary library of oligos even harder to construct than for DNA-PAINT.

[0008] More recently, a new version of DNA-PAINT called Resolution Enhancement for Sequential Imaging (RESI) has been developed that enables Angstrom-level resolution of labeled epitopes in a fixed biological sample. RESI works by solving the assignment problem for individual binding events, allowing reliable assignment of independent binding events in a single cycle to the exact same probe labeling a target epitope. This is achieved by applying a mixture of antibody probes containing the same antibody clone (specific for a single epitope) but different, orthogonal oligo sequences. This reduces the risk in a single round of imaging of encountering two epitopes labeled with antibodies containing the same oligo within the resolution limit achievable for the individual binding events for DNA-PAINT of roughly 20 nm. As individual binding events can now be reliably combined, the theoretically achievable accuracy of RESI is unlimited. Combination of >10 binding events was shown to already allow for Angstrom-level localization accuracy. RESI is the only widefield fluorescence approach capable of Angstrom resolution in fixed biological samples, but it comes at the twin costs of complexity and time.

[0009] A scanning-based approach to super-resolution based on combining stimulated emission depletion (STED) microscopy (Hell & Wichman, Optics Letters 19, 780, 1994; Klar & Hell, Optics Letters 24, 954, 1999) with STORM called MINFLUX has also beendemonstrated for Angstrom-level localization of single fluorophores within a biological sample (Balzarotti et al., Science 355, 606, 2017). Due to the scanning strategy of MINFLUX, only a single fluorophore can be localized at a time, hampering its speed of acquisition in comparison to the highly parallelized detection of the above widefield approaches to SMLM. MINFLUX is also not immediately adaptable for serial super-resolution of multiple markers in the same sample. The compatibility of MINFLUX with live cell acquisition is however a unique aspect of the technique and its main advantage, allowing for real-time acquisition of Angstrom-level trajectories of proteins within live cells. But such live cell applications lie outside the scope of the current patent, which addresses only SMLM of fixed biological samples.

[0010] For all of the above methods, the highest numerical aperture (N.A.) objectives based on oil immersion are typically employed to collect as much light as possible from each probe. For SMLM methods requiring the samples to be immersed in an aqueous buffer, the mismatch between the refractive indices of the oil and water leads to a considerable distortion of the point spread function (PSF) even at micrometer distances from the coverslip glass that also varies as a function of distance. To avoid these systematic effects on the SMLM localization analysis, observations very close to the glass interface are required, which for example is automatically satisfied with TIRF microscopy.

[0011] For all of the above approaches based on antibody staining, there is no possibility to discriminate the specific binding of the antibody to its target epitopes in the fixed biological sample from its non-specific binding background. The percentage of target epitopes successfully labeled by the antibody is also unknown. Blocking of the sample (e.g. using bovine serum albumin) can help to reduce (but not remove completely) the non-specific background, but too much blocking can also block the target epitopes and reduce specific labeling as well. For SMLM based on antibodies, the sample is typically blocked and then incubated with the antibody probe at a specific titer and for a specific duration of time considered optimal to reach as many epitopes as possible (with the exact labeling percentage unknown), while at the same time minimizing, but not completely removing, any remaining non-specific binding background. SMLM acquisition (whether STORM or DNA-PAINT or PALM) then proceeds, localizing indiscriminately specifically bound and non-specifically bound antibodies throughout the sample, with the final labeling percentage of the target epitope not quantifiable.

[0012] There is no currently available method for efficient, serial, distortion-free, detection-bias-free, Angstrom-level SMLM of arbitrarily many different epitopes in fixed biological samples at few-micron-level distances from the coverslip glass based on conventional binders, such as primary or secondary antibodies, labeled with standard fluorophores and that additionally allows for quantification of the labeling percentage of the target epitope as well as the ratio of specific binding to non-specific background.OBJECT OF THE INVENTION

[0013] In the following, a method called “Conventional Antibody Localization Microscopy” (CALM) is described that allows for SMLM down to Angstrom-level resolution of fluorophore-labeled binders (such as primary or secondary antibodies) that bind stably or semi-stably to target epitopes in a fixed biological sample.

[0014] In general, the invention is directed to a method for detecting target epitopes in a fixed biological sample with probes comprising an antigen binding moiety and a fluorescent moiety wherein the antigen binding moiety is capable of binding to at least a part of the target epitopes, characterized by the steps a. incubating the fixed biological sample for a specified time interval with a staining buffer containing the probes at a specific concentration to obtain a sparse labeling of the fixed biological sample with at least one isolated and therefore localizable bound probe, b. removing the staining buffer from the fixed biological sample, c. illuminating the fixed biological sample with a first light source and detecting the emission radiation from the bound probes in a first detection channel as a set of images formed on a first detector, d. repeating steps a to c.

[0015] Preferable, the term “incubating the fixed biological sample for a specified time interval with a staining buffer containing the probes at a specific concentration” refers to a dose of the probes i.e. a concentration over time.

[0016] Accordingly, object of the invention is a method for detecting target epitopes on a fixed biological sample with fluorescent probes comprising an antigen binding moiety and a fluorescent moiety, wherein the antigen binding moiety is capable of binding to at least a part of the target epitopes, characterized by the stepsa. incubating the fixed biological sample with a staining buffer comprising the probes wherein the dose d of the probes satisfies with F equal to the fraction of target epitopes notbound by probes, konthe on-rate constant, and log(O) = -oo, b. removing the staining buffer from the fixed biological sample, c. illuminating the fixed biological sample with a first light source and detecting the emission radiation from the probes bound to target epitopes in a first detection channel as a set of images formed on a first detector, d. repeating steps a to c.

[0017] Preferably, step c is conducted by illuminating the fixed biological sample with a first light source and detecting the emission radiation from the probes sparsely bound to target epitopes in a first detection channel as a set of images formed on a first detector for single molecule localization microscopy (SMLM).

[0018] Single molecule localization microscopy (SMLM) as used in the present invention comprises processes like for example CALM, PhotoActivated Localization Microscopy (PALM), STochastic Optical Reconstruction Microscopy (STORM), dSTORM, Point Accumulation for Imaging in Nanoscale Topography (PAINT), DNA-PAINT, IRIS, Fluorogemc DNA-PAINT, REST

[0019] The term “antigen binding moiety” refers to full antibodies, antibody-derived fragments, or molecules with antibody-like properties (antibody mimetics) recognizing for example pMHC molecules or ligand-bound MHC-like molecules. Antibody-derived fragments and molecules with antibody-like properties comprise, for example, Fab, Fab', F(ab')2, sdAb, scFv, di-scFv, scFv-Fc, nanobodies, DARPins, Anticalins, and monobodies

[0020] The term “target epitopes “ refers to any molecule on the fixed biological sample that can be bound by the antigen binding moieties, especially molecules expressed on a cell surface or DNA / RNA molecules.

[0021] The term “fraction of target epitopes not bound by probes” refers to so called “free target epitopes” which is controlled by the dose of probes as defined by the present invention.

[0022] Preferable, step b is performed by removal of the free probe from the sample leaving only the bound probe.

[0023] Preferable, step c is performed by acquisition of a sequence of images of the probes in one or more lateral locations of the sample, stopping acquisition in each location only when all (or most) probes have been either photobleached or have fallen off of the targeted epitopes.

[0024] Preferable step d is performed by the repetition of steps a-d as many times as desired (e.g. until a certain labeling percentage of the target epitope is reached), like 2 to 10000 times.

[0025] In the above, a primary binder probe could be a primary antibody. A secondary binder probe could be a secondary antibody that recognizes a primary antibody, or, for example, a fluorophore-labeled oligo (imager strand) that binds in a stable or semi-stable fashion to its complementary oligo target conjugated to a primary antibody.

[0026] Object of the invention is therefore a method for SMLM to generate a superresolution image of a fixed biological sample using conventional fluorophore-labeled binder probes that are applied in a gradual, stepwise manner to the sample, with each step specifically consisting of applying a low dose of the probe to the sample to achieve a sparse labeling, washing the free probe from the sample, acquiring and correctly assigning all emitted light to each separate probe position, and ending the acquisition only when all (or most) probes have been irreversibly photobleached, to both achieve optimal localization precision (the acquired signal from each probe is limited only be photobleaching) and to generate a clean slate for the next staining step. Significantly, the upper limit on the dose of probes used for CALM ensures that no more than 10% of the accessible epitopes are reached in a given step (see the detailed description below). This is far lower than the typical saturating single dose used for conventional staining of a sample with an antibody (e.g. for immunofluorescence or for other SMLM techniques, including serial staining approaches for SMLM), for which roughly a majority (>50%) of the accessible epitopes are bound. As the free probe is washed from the sample before image acquisition, probes with low affinity can be applied at high concentration with no adverse effects (in contrast to PAINT approaches). As well, the lack of a free probe background implies that URF is no longer required (as for PAINT), with conventional widefield epifluorescence microscopy possible to use, allowing observation of the sample to several micron distances from the coverslip glass.

[0027] Serial super-resolution using CALM can be performed after conclusion of the above steps a to d for a particular probe by simply repeating steps a to d for a different probe.The complete photobleaching (or unbinding) of the previous probe implies that it will not contribute to the imaging of the next probe.

[0028] The residence time (specifically, half-life) of the probes bound to their target epitopes should be roughly a factor of a few times longer than the washing timescale. With a microfluidic setup, washing of a microscopic field of view within the sample can be performed on a timescale of milliseconds to seconds. The residence time of the probes should therefore preferably be on the seconds or greater timescale. Also preferable would be a residence time that is longer than the photobleaching timescale, which is also on the order of milliseconds to seconds for typical illumination powers, in order to achieve maximum localization precision from each bound probe. As binding stability need only be on the seconds timescale, many other types of binders beyond conventional antibodies would be possible to observe with CALM as well, including less stable yet specific binders such as Fabs, scFvs, nanobodies, CDR3 peptides, physiological protein domains, physiological peptides, etc.

[0029] If the residence time of the probes is significantly longer than the complete process constituting multiple steps (stable CALM, or sCALM), then all probes, once bound, will remain bound to their specific epitopes throughout the complete process. With each step in the process, new probes will then only be able to bind to the remaining free epitopes. A gradual increase in probe concentration and / or incubation time should generally be employed over the entire process to ensure an optimal number of new probes are bound in each step. The process can be stopped when a certain percentage of epitopes (e.g. 50% or 95% or 99.9%) has been localized (as determined from the on-rate constant as estimated from the binding process, see derivation given below). This would allow for reliable percentage-wise assessment of the available epitopes that were so far successfully labeled (labeling efficiency), and, as a consequence, reliable inference of the total amount of accessible epitopes contained in the particular field of view of the sample.

[0030] If the probes reside longer than the second timescale but shorter than the entire process constituting the multiple steps (semi-stable CALM, or ssCALM), then there is a chance that some epitopes will be super-resolved in more than one step, which might seem to lead to a bias or over-counting (similar to PAINT-based approaches). However, for CALM, this over-counting effect can be removed, as the achieved resolution of each individual probe in a given step (down to few Angstrom lateral resolution and few nm axial resolution) is muchhigher than for PAINT-based approaches, allowing localizations obtained across multiple steps to be assigned to the exact same epitope position. This reliability of assignment also means that the signal collected across multiple steps can be directly combined for even greater localization precision of individual probe positions. As the process can be continued indefinitely for semi-stable probes, this implies an unlimited localization precision is possible with ssCALM.

[0031] In case of a high density of probes in a given step, two or more probes may overlap, making it difficult to separately resolve the individual probes. At high labeling density, the preferred probes to use for CALM would be those having a strict stoichiometric degree-of-labeling (DOL) of a single fluorophore to each probe (DOL=1). As the probes will be bleached in a step-wise fashion, this, along with the total intensity of a particular region of overlap, can be used to identify the number and positions of the multiple binders. If multiple fluorophores are present on each probe (DOL>1), stepwise bleaching would not be as helpful. For DOL>1, therefore, truly sparse detection with a small probability of overlap in each step would be more important, requiring an overall longer acquisition (number of steps) to superresolve the same desired percentage of total epitopes.

[0032] Multi-color SMLM using CALM can easily be achieved for fluorophores observed across different channels by acquiring the sequence of images in step c for the probes excited at the longest wavelength first and then the next longest wavelength, etc., preventing photobleaching of probes excited at shorter wavelength before they can be acquired. Roughly five different fluorophores in distinct emission channels spanning UV to optical to near IR wavelengths can be accommodated in this way.

[0033] Ratiometric SMLM, for which a single illumination wavelength is typically employed with the emission light from two or more spectrally distinct fluorophores split into two or more simultaneously detected emission channels on separate detectors, could also allow for more than one type of probe to be applied to the sample in a single step with increased relative positional accuracy over the above-described multi-color method based on detection in separate channels (e.g. Zhang et al., Nature Methods 12, 935, 2015). However, the benefits of ratiometric detection (parallel detection of distinct epitopes at higher relative spatial accuracy due to their simultaneous detection in the same channels) must be weighed against the optimal overall sparsity of total probe labels, which e.g. for five distinct probes may require a five-fold reduction in their individual densities and consequently a five-foldslower acquisition. Ratiometric detection, while providing a more reliable spatial accuracy, does not necessarily imply a significantly higher acquisition speed compared to detection is isolated channels (multi-color method) or even consecutive acquisition of each separate probe in different CALM processes.

[0034] CALM comes with several benefits in comparison to standard approaches to SMLM.

[0035] In contrast to PALM, use of a photo-activatable fluorophore — whether expressed as a fusion to a target protein (e.g. pa-GFP) or conjugated to an antibody (photo- activatable small molecule dye) — is unnecessary for CALM, as CALM can be applied to arbitrary standard fluorophores.

[0036] In contrast to STORM, all fluorophores from a single probe can be reliably assigned to single target epitopes for high-precision localization within a given step (reaching down to Angstrom-level precision), instead of the photon signal being widely distributed temporally as discrete blinking events. For STORM, the localization precision of individual blinking events arising from a single target epitope is typically insufficient to reliably assign all of the temporally distributed blinking events to the correct target epitope. The detection bias of STORM due to the statistics of blinking is also not present in CALM.

[0037] In contrast to PAINT, there is no free probe background for CALM as the free probe is washed from the sample before imaging. (While fluorogenic DNA-PAINT can reduce the free probe background by an order of magnitude, it does not eliminate it completely.) Use of an optical sectioning approach such as TIRF, spinning disk, or light sheet microscopy is therefore not required, with conventional widefield epifluorescence microscopy detection sufficient. Optical sectioning, while not necessary for CALM, can nevertheless still improve the precision in localization for all SMLM approaches by removing systematic errors generated by out-of-focus light from probes present on the sample either above or below the focal plane.

[0038] In contrast to DNA-PAINT, there is no need to conjugate oligos to conventional antibodies. Conventional antibody probes used routinely for immunofluorescence can be directly employed for CALM.

[0039] In contrast to RESI, a single localization event is sufficient in CALM to achieve Angstrom-level resolution.

[0040] In contrast to IRIS, there is no restriction to highly transient probes having half-lifes less than roughly a few seconds, which excludes many classes of standard binders. For CALM, probes should preferably have a half-life greater than a few seconds, which includes standard antibodies, Fabs, scFvs, nanobodies, as well as many potential binders based on physiological protein domains or peptides (that would not be as suitable for IRIS due to a too high half-life).

[0041] In contrast to standard SMLM approaches, it is not essential to use the highest N.A. objectives based on oil immersion. The extremely high detection efficiency of CALM implies that lower N.A. objectives based on water immersion could be seriously considered, as the loss in the already very high resolution would not be so worrisome. Use of water immersion objectives with fixed biological samples immersed in aqueous buffers (the latter required for PAINT) implies less distortion effects at few micron depth in the sample than for the oil / water refractive index mismatch presented by an oil-immersion objective.

[0042] In contrast to other antibody-based SMLM approaches that are unable to discriminate specific binding from non-specific binding, with CALM it is possible to quantify the specific binding versus non-specific binding of the fluorescent binder probe, yielding an estimate of the percentage of the target epitope that has so far been labeled as well as the ratio of signal to non-specific background labeling (see derivation below). This is possible as the CALM process is akin to a titration process, allowing discrimination of the specific signal that gradually reaches saturation from the non-specific binding that remains unsaturated.BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 shows a unidirectional microfluidic setup for CALM with separate steps of (A) staining, (B) washing, and (C) localization. Here, a series of consecutive staining buffers is required (stain / ., stai +i, etc.).

[0044] Figure 2 shows a bidirectional microfluidic setup for CALM for which a single staining buffer (staim) is applied to the sample in a repeated fashion in a cycle of forward and reverse flow. (A) First staining step. (B) Forward flow to washing step. (C) Forward flow to localization step. (D) Reverse flow to washing step. (E) Reverse flow to second staining step.

[0045] Figure 3 shows a unidirectional microfluidic setup for CALM for which a single staim buffer is applied to the sample in a repeated fashion but always in the same flow direction through use of a storage loop. (A) Microfluidic setup including storage loop. (B)Process steps, indicating the opening (“O”) and closing (“C”) of valves, required to cycle the staim buffer between the sample location and the storage loop.

[0046] Figure 4 shows two steps of the CALM approach for an APC-labeled antibody against actin (β-Actin Antibody, anti-human / mouse / rat, APC, REAfinity™, 130-120-278, Miltenyi Biotec ).DESCRIPTION OF THE INVENTION

[0047] Detailed here is a method called CALM for SMLM based on conventional fluorophore-labeled molecular probes (e.g. antibodies) that are applied in repeated steps of sparse labeling, washing, and imaging / photobleaching to a fixed biological sample.

[0048] In the drawings the following reference numbers are used to refer to the following features. Similar reference numbers are used in the various figures to refer to components that serve a similar or identical function.001 Sample chamber002 Sample chamber width003 Sample chamber height004 Forward fluid flow direction005 Fixed biological sample006 Microscope objective007 Coordinate system008 Buffer containing fluorescent molecular probe009 Angle of light collection010 Leftmost extent of objective light collection cone011 Rightmost extent of objective light collection cone012 Leftmost boundary of the localization region in the sample013 Rightmost boundary of the localization region in the sample014 Leftmost extent of light collection cone when the obj ective is positioned at the leftmost boundary of the localization region015 Rightmost extent of light collection cone when the objective is positioned at the rightmost boundary of the localization region016 Reverse fluid flow direction017 Storage loop018 Microfluidic valve019 Microcolony consisting of four PFA-fixed HeLa cells020 Region of interest021 Single isolated fluorophore in step 1 of CALM process022 Single isolated fluorophore in step 2 of CALM process

[0049] In Figure 1, a schematic of a preferred microfluidic setup under laminar flow conditions for CALM is displayed. A fixed biological sample (005) is placed in a sample chamber (001) of specific width (002) along the x-axis (007), which coincides with the direction of flow (004), and of specific height (003) along the y-axis (007) above the biological sample. The sample is observed with through a microscope objective (006). Three specific stages corresponding to a single step k in the CALM process are shown in panels A, B, and C. In A, the sample is incubated in the staim buffer (008) containing the molecular binder probe at a specific concentration. The staining can take place under continual flow or the flow can be stopped for a specific time interval. The total incubation time for staim on the sample then corresponds to the length of the staim buffer along the x axis (007) divided by the constant flow speed plus the time interval during which the flow was stopped. In B, the sample is washed by flowing the wash / , buffer over the sample, with the efficiency of washing depending on both the length of the wash / , buffer along the x axis and the total time required to flow the wash / , buffer over the sample. In C, a sequence of images is acquired of the bound fluorescent probe until all (or most) of the probes in one or more laterally displaced regions of the sample are photobleached. The extent of the imaging region is determined in part by the collection angle (009) of the objective (determined by the numerical aperture, N.A.), with the leftmost (010) and rightmost (Oi l) boundaries of the light cone shown for the displayed position of the objective (here, the leftmost and rightmost boundaries refer, respectively, to the separate light cones determined at the leftmost pixel and rightmost pixel positions of the detector field of view in the sample). In order to avoid collecting fluorescence light from either of the staim+i or staim buffers, the sample should not be moved further than the left (012) or right (013) boundaries of the localizes region along the x axis (007). When the objective is at the left edge (012) of the localizes region, the leftmost boundary of the light cone (014) should not intersect the staim+i buffer. Similarly, when the objective is positionedat the right edge (013) of the localizes region, the rightmost boundary of the light cone (015) should not intersect the staim buffer.

[0050] In Figure 2, a microfluidic setup similar to Figure 1 is shown with the additional possibility for reversal of the fluid flow direction (016). The exact same staim buffer (008) containing probes can then be applied to the fixed biological sample (005) in a repeated fashion. Specifically, in panel A, the staim buffer (008) containing the probes at a specific concentration is flowed onto the sample in the forward direction (as in Figure 1). In panel B, the washi buffer is flowed onto the sample in the forward direction. In panel C, the localization! buffer is flowed on the sample and isolated probes are then detected for localization. In panel D, the fluid flow is reversed (016) to place the sample again in the washi buffer (though washing here is unnecessary). In panel E, the reversed fluid flow is continued in order to place the staim buffer on the fixed biological sample again, returning the system to the same state as in panel A. In this approach, therefore, the same staim buffer can be applied and reapplied to the sample as many times as are necessary to localize a specific fraction of the targeted epitope, which allows performance of the CALM process using far less probe than in the unidirectional flow depicted in Figure 1.

[0051] In Figure 3, a unidirectional microfluidic setup that allows for the same staim buffer to be applied to the sample in a repeated fashion but also in the same flow direction (to avoid perturbing the sample by reversal of flow) for each step is shown. (A) Schematic of the microfluidic setup containing a storage loop (017), several valves (018), two inlets, and two outlets. (B) Microfluidic steps that allow for repeated staining of the sample with the staim buffer via opening (“O”) or closing (“X”) of the valves and activation of the relevant flow through inleti or inlet2.

[0052] Figure 4 shows the incubation of the sample at a low concentration of probe (100* lower than recommended for conventional IF) and for a short duration (1 min vs. 10 min for conventional IF) are shown. Individual detected spots in one step can be superresolved and combined (following registration based on reference to a transmission image) to form a super-resolution image. (A) Full image of frame 1 of the image sequence acquired after first staining step. The cropped region shown in panel A is displayed for (B) frame 1, (C) frame 6, (D) frame 42, and (E) frame 155. (F) Full image of frame 1 of the image sequence acquired after second staining step. The cropped region shown in panel F is displayed for (G) frame 1, (H) frame 6, (I) frame 42, and (J) frame 155. (K) Image of the same region shown inpanels A and F but now obtained after staining the sample with the same probe but according to the manufacturer's recommendation for conventional IF. (L) Cropped region shown in panel K.

[0053] The CALM process in Figure 4 is carried out on a cluster of four PFA-fixed HeLa cells (019). A 1:5000 dilution of a beta- Actin antibody conjugated to the dye APC (130- 120-278, Miltenyi Biotec B.V. & Co. KG) was applied to the cells for 1 min (1000-fold lower dose than the dose recommended by the manufacturer for conventional IF). In A, the full illuminated field of view is shown for Step 1 of the CALM process for the first 100-ms frame acquired after activating the 640 nm laser. In B - E, a zoomed in view of the region of interest (020) in panel A is shown respectively for the frames recorded from 0.0 - 0.1 s (first frame, as in A), 0.5 - 0.6 s, 4.1 - 4.2 s, and 15.4 - 15.5 s. For the single isolated APC fluorophore (021), 109,403 photons, corresponding to an improved localization resolution (based purely on photon counting statistics) down to 0.61 nm (determined by dividing the conventional 200 nm resolution by the square root of the number of detected photons), were recorded before its photobleaching. Photon counts were determined by integrating the intensity over the pixels in the boxed region (021) over the entire stack of images (the gain of 1.0 for the detector implies intensity values reflect photon numbers) and then subtracting the integrated intensity of a similar boxed region over an adjacent fluorophore-free region. In panels F - J, similar images are shown for step 2, with 343,169 (0.34 nm localization resolution) photons recorded for the localized fluorophore in the boxed region (022). In K, a conventional IF image was acquired of the sample using the same antibody and following the manufacturer's recommended concentration and incubation time (10 min). In L, the same zoomed in region of K as that shown in panels B - E and G - J is shown. The actin cytoskeletal network in the interior of these HeLa cells was disrupted by the fixation process, leading to very short filaments (panel K) instead of the expected continuous network.

[0054] The most important quantity for CALM is the dose of the probe on the sample in each step k, dk, which is equal to the probe concentration, at, times the incubation time, tk.

[0055] For stably binding probes that remain bound to their targets through the entire cyclic process, we can neglect the off-rate constant for unbinding (stable CALM, sCALM). The fractions of free and bound epitope are then only determined by the on-rate constant and the applied dose of the probe. Iterative staining steps are most easily expressed in terms of the exponential reduction of the fraction of free epitope, fk, after each step k, withContinuing the expansion down to the first step, the free fraction following step k is thenwith f0= 1 assumed and the accumulated dose of probe on the sample, Dk, defined as Dk= Yi=i dt- For a series of concentrations, at, applied in succession over the respective time intervals, tt= At, the sum defining the accumulated dose can be written as an integral in the standard way: with a(t) giving the instantaneousapplied concentration as a function of time and T = For = 1 in the above,we are accounting only for the target epitopes in the fixed sample that are potentially reachable by the probe: Some epitopes will not be reachable due to their direct fixation or due to the impermeability of the fixed sample (e.g. caused by lipids or the fixation matrix), hindering the probe’s access to the epitope. The fraction of labeling of the accessible epitopes in a given step, k, is defined as lkwith lby definition. Expressing lkas the fractional decrease of free probe from step k — 1 to step k gives:Setting an upper limit of lk< L, implying e.g. for L = 0.1 that no more than 10% of the total accessible epitopes is labeled in the step k, gives the following limit on the dose dkof:For L -» fk-1, the argument of the logarithm goes to 0, giving dk< co, implying no limit to the dose that can be applied. The bound fraction, bk, at each step k is simplyIf the immunostaining incubation time is kept constant in each step (ti= tc), then:with the summed concentration of probe applied to the sample defined asthis case, the iterative immunostaining at step k would be equivalent to a single staining stepof duration tcat concentration Ak. Alternatively, if the concentration of probe is held constant for each step (aiac), then:with the total integrated time for which the probe was applied to the sample defined as Here, the iterative immunostaining at step k would be equivalentto a single staining step at concentration acfor the total integrated time Tk.

[0056] In the above, the on-rate constant, kon, was assumed known. However, if the on-rate constant is not known, it can easily be inferred from the image sequence in the following manner. The dose, dk, for each staining step k is known (as both the probe concentration and incubation time are set by the experimenter). We need therefore simply consider the number of newly bound probes, nk, in each step k as observed in the corresponding image sequence for that step and the total remaining free epitope following step k — 1, which isstep binding to the remaining total number of free sites E(1 — bk-1i) is described by a binomial distribution with binding probability for each site given byIn an exact sense, the number of newly bound epitopes, nk, follows a binomial distribution with nksuccesses and Fk_i — nkfailures, with the latter specified by the probability qk= 1 ~ Pk = exp(-kconaktk) orForPk « 1 and for nk> 10, which are both typical assumptions for CALM (only failing at very high binding saturation of the targeted epitopes), the probability distribution for the amount of newly bound epitope in step k can be approximated as a Gaussian: 9k(nk; kon)Based on the known dkand observed nk, we can fit for E and konby maximizing the likelihood, which is the product of the individual Gaussian probabilities:where S corresponds to the total number of steps. Maximizing the likelihood with respect to E and konis identical to maximizing the log of the likelihood:The pair of values that maximize the likelihood will be written as {E) and kon). Once the on- rate constant has been estimated, the percentage of available epitope, Ps= 100bs-, reached by the final step S can be determined:Ps= 100(1 - exp(— <kon)Ds)]The dose, Ds, needed to reach a certain target percentage by the final step S is then:A more exact calculation could be used based on maximizing the likelihood for the full binomial distribution. It is well known that maximization of the likelihood generates a slightly biased estimate. Maximization of the fidelity has been shown to generate a less biased estimate than the likelihood for many standard probability distributions (Walther et al. 2011, Molecular BioSystems 7, 322, 2011; Kinkhabwala 2013, arXiv:1301.5186) and would therefore be the preferred option. The fidelity statistic is based on the cumulative distribution and could be applied to the respective cumulative distributions for each step of the process derived from either the approximate Gaussian probability distributions or the more exact binomial probability distributions. For the collection of approximate Gaussian probability distributions, the joint fidelity, f, is defined as the sum over the individual fidelities defined for each individual distribution, or:with Gfc(nk; E, kon) = the cumulative distribution of the Gaussian(identical to the standard error function).

[0057] If there is also non-specific binding of the probe in each step, this can typically be assumed to contribute in a linear fashion with the applied dose of the probe. This linear relation is due to the fact that non-specific binding sites on the sample are numerous but also characterized by a low on-rate constant, implying that these sites remain far from saturation over the range of doses that lead to saturation of the specific signal. A linear-with-dose nonspecific background in the above global analysis could be accommodated as an additive model component at each step contributing an additional mean number of events equal to a global multiplicative parameter times the applied dose for that step.

[0058] For binders that remain bound on the sample through the washing and localization phases of a single step, but not necessarily for the entire CALM process, a single epitope can be therefore be bound one or more times in other steps (semi-stable CALM, ssCALM). As the off-rate constant now plays a significant role, the above mathematical treatment for sCALM, therefore, no longer directly applies. However, a more complicated global analysis that keeps track of the revisitations of specific epitope sites can be formulated with the off-rate constant a new global parameter. An additive component for a linear-with- dose non-specific background could also be included as described immediately above. While analysis of the binding reaction and inference of total epitope number is more complicated, ssCALM has the advantage that individual epitopes can be localized in more than one step, further increasing their localization accuracy. As the localization accuracy obtained within a single step (see Figure 4) is already sufficient to achieve down to few Angstrom resolution, individual localizations obtained across separate steps of the process can robustly be assigned to the exact same epitope. For ssCALM, therefore, there is no limit to the obtainable localization accuracy. Based on the assumption that individual localizations across distinct steps in the CALM process can be reliably assigned to the same epitope, it is possible to infer the total epitope number as well.

[0059] For CALM, a photobleaching movie is acquired at each step that can be analyzed using a global analysis based on maximum likelihood in the following way. Ideally, a complete detector noise model (offset, readout noise, dark count rate, gain) has been determined beforehand, allowing discrimination of true signal from detector effects and assignment of a Gaussian error bar to each observed pixel intensity. To account for out-of-focus blurred spots, a low-pass frequency filter can be applied to the entire movie before localization analysis with spatial frequency cutoff chosen significantly larger than the resolution limit to avoid loss of localization information for individual probes. This generates an image of the out-of-focus blur for each frame. Then, the movie is run in a reversed sequence with a single probe assigned to each isolated object detected in each new frame. Overlapping objects from frame to frame are assigned the same probe identity unless the shape of the probe is significantly altered indicating the presence of an additional fluorophore, for which a displaced position and corresponding intensity distribution are assigned. The individual localizations in each frame are then used as initial parameter values for maximization of the likelihood over the stack, carried out as a global analysis. Here, the global parameters are the fixed 3D positions of each probe and the local parameters are the probe intensities in each frame (to account for variable blinking and photobleaching). The individual probe distributions are scaled by the local intensities for each frame and are added to the image of the out-of-focus blur to determine the full model intensities for each pixel and the likelihood. The global and local parameters are then nonlinearly optimized to find the optimal solution for the 3D positions of all probes. Of note, maximization of the likelihood based on the Gaussian error bar in each pixel is equivalent to a weighted least squares minimization with the weights equal to the Gaussian variance in each pixel.

[0060] In one embodiment, the primary probe could be comprised of a first moiety capable of binding a target epitope and a second moiety. In this case, the secondary probe could be comprised of a fluorophore- labeled third moiety capable of binding in a stable or semi-stable fashion (>10 sec half-life) to the second moiety. As specific examples, the second moiety and third moiety could be comprised of a pair of complementary oligonucleotides, or they could also be comprised of a pair of amino acid polymers such as peptides or protein domains.

[0061] In another embodiment, delivery of the staining, washing, and imaging buffers to the sample could be carried out by manual pipetting. Manual pipetting was used to generate the experimental images shown in Figure 4.

[0062] In another embodiment, a unidirectional microfluidic setup consisting of a sample chamber with an inlet and an outlet (as depicted in Figure 1) is used to deliver and remove the different buffers from the sample. Individual staining steps (stain / . in Figure 1) are delivered in a sequential, unidirectional fashion to the fixed biological sample.

[0063] In another embodiment, a bidirectional microfluidic setup is used to reduce the total amount of probe required for the CALM process (Figure 2). The bidirectional setup would be similar to the unidirectional one shown in Figure 1 aside from the ability to reverse the flow direction (polarity of the pump). Here, specifically, only a single staining step, staim, is delivered to the sample and removed from the sample repeatedly by employing a forward flow followed by a reverse flow for each cycle. In this embodiment, the same washing and localization buffers would also be used for each cycle.

[0064] In another embodiment, to maintain a single flow direction over the sample as well as the above advantage of a single staining step applied multiple times to the sample, a microfluidic setup is employed (Figure 3A) consisting of two inlets, two outlets, a storage loop (017), and six valves (018). Through the steps shown in Figure 3B, a coordinated opening (“O”) and closing (“X”) of the valves in each step (and activation of the relevant flows through either inleti or inlet2) would allow the same staim buffer to be applied to the sample in a unidirectional sense and in a repeated fashion. Specifically, in Step 1, the staim buffer is for the first time delivered to the sample from afar. In Step 2, the staim buffer is transported to the storage loop (017). In Step 3, the sample is washed, after which the localization step can be performed. In Step 4, the staim buffer is transported from the storage loop (017) and back to the sample. Steps 2 - 4 can be repeated as many times as desired. In Step 5, the sample is removed to the waste. At this point, a new probe could be applied to the sample from afar, repeating the process starting from Step 1.

[0065] In another embodiment, a cylindrical lens (for astigmatic imaging, Huang et al. Science 319, 810, 2008) or diffraction element (for SELFI, Linares-Loy ez et al., Frontiers in Physics 7, 68, 2019) is used to increase axial resolution. In both approaches, the detected light from the probe is distorted in a different way if the particle is located above the focal or below the focal plane, with the distance above or below the plane determined by the degree of distortion (e.g. by the ratio of the major and minor axes of the elliptical distortion created in astigmatic imaging). Improvement in axial resolution for typical SMLM recordings is limited to roughly 50 nm by the few hundred to few thousand photons that can be collected per individual binding event for PAINT, or individual blinking event for dSTORM. As CALM allows for the reliable detection of hundreds of thousands of photons from a single probe, axial resolution down to roughly a few nm should be achievable.

[0066] In another embodiment, a pulsed laser is used for the illumination to reduce the rate of fluorophore photobleaching and thereby increase further the amount of photons that can be collected before irreversible photobleaching (Donnert et al., Nature Methods 4, 81, 2007). Photobleaching is thought to primarily occur through photoionization of the pool of fluorophores that end up in a long-lived triplet state (< 1 ps) following an initial photoexcitation and low probability transition to the triplet state. A pulsed laser with spacing on the roughly >1 ps timescale would allow for the triplet-state fluorophores to de-excite, preventing their photoionization and irreversible photobleaching. Donnert et al. have reported a 5 - 25 x improvement in total fluorescence intensity before photobleaching (for the dye Atto 532) using such a low pulse repetition rate.

[0067] In another embodiment, oxygen scavengers are added to the localization buffer in order to both decrease the photobleaching rate as well as to reduce oxidative damage resulting as a side effect from the high illumination of the fixed biological sample.

[0068] In another embodiment, a buffer with specific redox potential for the fluorophore is used for the localization step in order to minimize photobleaching and thereby maximize the total number of photons detected before photobleaching.

[0069] In another embodiment, a spatial light modulator (SLM) in a conjugate plane is used to direct the excitation light only to the positions in the sample that contain probes to be localized. Such a strategy reduces the total irradiation of the sample, thereby reducing overall photodamage to the sample (Henriques et al., Biopolymers 95, 322, 2011) as well as reducing the excitation level of out-of-focus probes to minimize their background contribution to the images of the in-focus probes. Here, a first image or images (to account for blinking of the probes) must be taken to identify the probe positions, which are then used as the basis for determining the specific SLM pattern that should be applied to all subsequent images for the given step.

[0070] In another embodiment, the SLM is a digital micromirror device (DMD) with two mirror angles for each pixel element (conjugate and non-conjugate) that is additionally used to direct the emission light from the probes onto a first detector (conjugate pixels of the DMD, which are oriented ) as well as the out-of-focus light onto an additional detector (nonconjugate pixels of the DMD) to create a programmable array microscope (PAM). The out-of- focus non-conjugate light collected on the additional detector can then be appropriately scaledand subtracted from the light detected on the first detector. Alternatively, the full information from both images can be retained for performing the localization analysis.

Claims

29Claims1. Method for detecting target epitopes on a fixed biological sample with fluorescent probes comprising an antigen binding moiety and a fluorescent moiety, wherein the antigen binding moiety is capable of binding to at least a part of the target epitopes, characterized by the steps a. incubating the fixed biological sample with a staining buffer comprising the probes wherein the dose d of the probes satisfies wjth F equal to the fraction of target epitopes notbound by probes, fconthe on-rate constant, and log(O) = -oo, b. removing the staining buffer from the fixed biological sample, c. illuminating the fixed biological sample with a first light source and detecting the emission radiation from the probes bound to target epitopes in a first detection channel as a set of images formed on a first detector, d. repeating steps a to c.

2. Method according to claim 1 characterized in illuminating the fixed biological sample with a first light source and detecting the emission radiation from the probes sparsely bound to target epitopes in a first detection channel as a set of images formed on a first detector for single molecule localization microscopy (SMLM).

3. Method according to claim 1 or 2 characterized in estimating the centers of the intensity distributions for the probes bound to the target epitopes that are observed over one or more images of the set of images to localize them.

4. Method according to any of claims 1 to 3 characterized by washing the fixed biological sample with a washing buffer after step b.

5. Method according to any of claims 1 to 4 characterized by incubating the fixed biological sample with a buffer containing oxygen scavengers before step c.

6. Method according to any of claims 1 to 5 characterized by halting the acquisition of the set of images when the emission radiation from at least one of the bound probes is no longer detected due to photobleaching.

7. Method according to claims 1 to 6 characterized by using a unidirectional flow to deliver and remove the buffers from the fixed biological sample.

8. Method according to claims 1 to 7 characterized by using a bidirectional flow to deliver the same staining buffer to the fixed biological sample using a cycle of forward flow followed by reverse flow.

9. Method according to claim 8 characterized by using a storage loop for temporary storage of a single staining buffer that is delivered to and removed from the fixed biological sample in a repeated unidirectional manner.

10. Method according to any of claims 1 to 9 characterized by providing in step a at least one additional fluorescent binder probe at a specific concentration, and in step c illuminating the sample with at least one second light source and detecting the emission radiation in at least one second detection channel.

11. Method according to any of claims 1 to 10 characterized by using a pulsed laser with specific repetition rate for the first light source.

12. Method according to any of claims 1 to 11 characterized by using a pulsed laser with specific repetition rate for the second light source.

13. Method according to any of claims 1 to 12 characterized by in step c splitting the emitted radiation into at least one third detection channel and detecting as a set of images on at least one second detector.

14. Method according to any of claims 1 to 13 characterized by repeating steps a to c for at least one additional fluorescent binder probe.

Images