Macroscopic Fluorescence Lifetime Imaging for Monitoring of Drug–Target Engagement

Abstract

Precision medicine promises to improve therapeutic efficacy while reducing adverse effects, especially in oncology. However, despite great progresses in recent years, precision medicine for cancer treatment is not always part of routine care. Indeed, the ability to specifically tailor therapies to distinct patient profiles requires still significant improvements in targeted therapy development as well as decreases in drug treatment failures. In this regard, preclinical animal research is fundamental to advance our understanding of tumor biology, and diagnostic and therapeutic response. Most importantly, the ability to measure drug–target engagement accurately in live and intact animals is critical in guiding the development and optimization of targeted therapy. However, a major limitation of preclinical molecular imaging modalities is their lack of capability to directly and quantitatively discriminate between drug accumulation and drug–target engagement at the pathological site. Recently, we have developed Macroscopic Fluorescence Lifetime Imaging (MFLI) as a unique feature of optical imaging to quantitate in vivo drug–target engagement. MFLI quantitatively reports on nanoscale interactions via lifetime-sensing of Förster Resonance Energy Transfer (FRET) in live, intact animals. Hence, MFLI FRET acts as a direct reporter of receptor dimerization and target engagement via the measurement of the fraction of labeled-donor entity undergoing binding to its respective receptor. MFLI is expected to greatly impact preclinical imaging and also adjacent fields such as image-guided surgery and drug development.

1. Introduction

Preclinical molecular imaging is commonly employed during early drug development and is an invaluable research tool to better understand the biological mechanisms of drug action and of acquisition of drug resistance. Due to their high sensitivity, nuclear (positron emission tomography, PET) and optical imaging (OI) are the two main techniques used to noninvasively assess the efficiency of drug delivery, efficacy, pharmacokinetics, and response in longitudinal studies [1–3]. Dynamic PET imaging provides the spatial and temporal distribution of radiotracers in living animals with a few restrictions [4]. However, PET imaging is limited to monitoring the local accumulation of a single targeted radiotracer. On the other hand, OI has become the alternative imaging tool most commonly used in preclinical studies [5–7]. Some advantages of OI are: (a) imaging simultaneously multiple markers that can monitor the molecular, structural, and physiological state of the tissue [8–10], (b) increasing number of exogenous probes or gene reporters [11–13], (c) diverse methodology using fluorescence intensity, spectral or lifetime-sensing in longitudinal studies [14–16].

In the case of antibody–drug conjugates, which have been developed as highly potent, targeted therapies against various cancers, co-localization of radiotracer/fluorescent probes with the pathological site does not guarantee interaction of the antibody with the target protein. Indeed, it is well recognized in microscopy that co-localization is not appropriate for detecting molecular interactions such as receptor–ligand target engagement. However, this interaction that reflects receptor engagement is essential to elicit a cellular response to eradicate the cancer cells [17]. Similarly to PET, OI allows to dynamically image the ligand-targeted tumor accumulation [18, 19] using multi-compartmental analysis [20, 21]. Nevertheless, it is difficult to discriminate between nonspecific accumulations from receptor-mediated tumor uptake without the use of invasive biochemical studies or rather complex multi-compartmental pharmacokinetics models [22–25].

Förster Resonance Energy Transfer (FRET) imaging can sense protein–protein interaction events at the distance of 2–10 nm, which is the range in which binding of antibodies/protein ligands to their respective receptors at the cell surface can occur. Fluorescence lifetime imaging (FLI) measures FRET events by determining the reduction of fluorescence lifetime of donor fluorophore when near to one or more acceptor fluorophores [26]. For example, when applied to ligand-receptor systems, FRET occurs when donor-labeled and acceptor-labeled ligands bind to dimerized/clustered receptors [27–29]. Hence, FLIM FRET acts as a direct reporter of receptor dimerization and target engagement via the measurement of the fraction of labeled-donor entity undergoing binding to its respective receptor [30, 31]. FLIM FRET has been confined to microscopic techniques, and its translation to small animal in vivo imaging preclinical research has been achieved only recently.

For preclinical studies, especially ones associated with the assessment of drug delivery and efficacy, it is imperative to be able to image the whole body of small animal models. Hence, the imaging approaches should be able to image a large field of view with high sensitivity. Additionally, biological investigations are greatly enhanced by the ability to image multiple biomarkers simultaneously. This can be done by multiplexing the fluorescence probes spectrally and via lifetime sensing. However, increasing the dimensionality of the data to be acquired leads to increase complexity in the imaging system and burden in acquisition time. Herein, we report on two systems we have developed to perform Macroscopic Fluorescence Lifetime Imaging (MFLI) in live animals. The first system is a multispectral imager that allows consecutive acquisition of spectral channels. It is based on an innovative time-resolved wide-field illumination strategy [32] and a time-gated intensified charge-coupled device (ICCD) camera to accurately perform MFLI on time-gated data sets [30, 33] The second system is capable of performing Hyperspectral Macroscopic Fluorescence Lifetime Imaging (HMFLI) for increased information content [34, 35]. This system is centered around the concept of single-pixel imaging coupled with time-resolved spectrophotometry. Both instrumental approaches have been validated in quantitative reporting on target-receptor interaction via in vivo FRET [26, 36, 37] and both can perform 2D and 3D tomographic multiplexed depth-resolved imaging [38–44].

Over the last few years, we have successfully validated MFLI-FRET in assessing tumor xenografts for receptor-targeted uptake using the gold standard ex vivo immunohistochemistry (IHC) analysis [45, 46]. We have used the transferrin receptor (TfR)–transferrin (Tf) complexes as a proof of principle for receptor–ligand binding in tumors since TfR is overexpressed in a variety of cancers and Tf is widely used as a ligand/carrier in targeted anti-cancer therapy. In vivo MFLI FRET in tumor xenografts has been attained using near-infrared (NIR)-labeled Tf molecules as FRET partners, since FRET only occurs when donor-labeled Tf and acceptor-labeled Tf bind with homodimeric TfR [31, 37, 46]. Hence, MFLI FRET acts as a direct transduction mechanism to quantify target engagement by determining the amplitude of the short lifetime component of the donor molecule (A1), i.e. FRET donor fraction (FD%). The experimental approach includes the intravenous injection of acceptor (A) fluorophore (Alexa Fluor 750; AF750)-probe and donor (D) fluorophore (Alexa Fluor 700; AF700)-probe conjugates at A:D ~ 2 ratio, followed by in vivo MFLI imaging. Monitoring the acceptor channel under donor excitation does not provide means to robustly monitor FRET in vivo using fluorescence intensity-based in vivo small animal imaging systems [47, 48], whereas lifetime-based MFLI imaging at the donor wavelength provides an accurate and robust quantitative approach. Strikingly, we have showed that MFLI-FRET measurements correlate with Tf binding to tumor cells but, not with TfR expression assessment. This is a critical issue in targeted drug therapy as ex vivo receptor profiling is the main method employed in tumor diagnosis and therapy despite that targeted drug delivery in vivo may not always correlate with target expression levels. Recently, MFLI FRET has been applied to clinically relevant biomarkers, such as HER2 [49], and to multiplexed longitudinal imaging [50] to evaluate drug response and efficacy.

Overall, our lifetime-based approach allows for the noninvasive visualization and quantification of FRET and measurement of protein–protein interactions in live, intact animals. Here, we describe the technological innovations and methodological developments that set the foundation to image in vivo and noninvasively target engagement in deep tissues in living animals.

2. Materials

2.1. Optical Components for ICCD-Gated MFLI

The ICCD-gated apparatus is configured in reflectance and transmission geometry [33]. The reflectance geometry is employed for subcutaneous tumor and fast imaging while transmission geometry is used when 3D imaging is required. For simplicity, we will focus on 2D imaging herein. In this setting, the system acquires both 2D intensity and time-resolved data. The following components are used in the arrangement [30, 31, 51]:

• Light Source:

  • Ti: Sapphire Mai Tai HP (Spectra-Physics, CA): 690–1040 nm emission, 100 fs width, 80 MHz repetition rate, Power range of 1 W to 3 W.
  • Power control module CCVA-TL-KT (Newport Optics, CA).
  • Fiber Coupler (F-91-C1-T with M-10X objective, Newport Optics, CA).
  • Multimode Fiber 400 μm core-diameter (P-400–10-VIS-NIR, Ocean Optics Inc., FL).
  • Collimator (HPUCO-25-690/950-M-35 AC-UNL, OZ Optics, Ontario, Canada).
  • Focus (62584, Edmund Optics, USA).
  • Optical Constant Fraction Discriminator (trigger generator, OCF-401–1, Becker & Hickl GmbH, Germany).

• Illumination DMD in reflectance geometry: Digital micro-mirror device (DMD, D2D module, Texas Instruments Inc).

• Dark Imaging Plate at f = 280 mm from the illumination DMD.

• ICCD Camera (Picostar HR, LaVision GmbH) equipped with High Rate Imager (HRI; Kentech Instruments, U.K.).

• S2+ Optics Module (Texas Instruments Inc.).

2.2. Optical Components for Hyperspectral MFLI

The HMFLI system is configured on a single pixel arrangement. The system enables the acquisition of 2D intensity and time-resolved data over 16 spectral channels, simultaneously. This computational imaging strategy requires to inverse solve the image plane based on one-dimensional measurements acquired through a setup assembled from the following components [34, 35]:

• Light Source:

  • Mai Tai HP (Spectra-Physics, CA): 690–1040 nm emission, 100 fs width, 80 MHz repetition rate, Power range of 1 W to 3 W.
  • Power control module CCVA-TL-KT (Newport Optics, CA).
  • Fiber Coupler (F-91-C1-T with M-10X objective, Newport Optics, CA).
  • Multimode Fiber 400 μm core-diameter (P-400-10-VIS-NIR, Ocean Optics Inc., FL).
  • Fiber-coupled light to collimator F220-SMA-B and then lens f = 25.4 mm. Arrangement coupled to illumination DMD device.
  • Optical Constant Fraction Discriminator (trigger generator, OCF-401-1, Becker & Hickl GmbH, Germany).

• Illumination DMD in reflectance geometry: Spatial Light Modulator PK101 (Optoma, CA) with removed collimating lenses.

• Dark Imaging Plate at f = 119 mm from the Illumination DMD.

• Mirror and external NIR complementary metal-oxide semiconductor (CMOS) Camera (CMOS: UI-5240CP-NIR-GL, IDS Gmbh, Germany; Lenses: HF12.5HA-1B or HF35HA-1B, FUJIFILM Corporation, Japan).

• Detection DMD, D4110 with NIR S2+ optics (Digital Light Innovations, TX).

• Relay Optics with lenses:

  • 3 Lenses (f = 30.0 mm).
  • 2 Lenses (f = 25.4 mm).
  • Integrator Rod (63–086, Hexagonal light pipe, Edmund Optics, NJ).
  • Lens (f = 40.0 mm).

• Fiber Bundle/Light guide.

• Spectrophotometer:

  • 2 Mirrors.
  • Grating (77412, 750 nm blaze, 1200 lines/mm, Newport Optics, CA).
• MW-FLIM Detector:

  • Multi-alkali time-resolved photomultiplier tube (PMT) (PML-16-C, Becker & Hickl GmbH, Germany).
  • SPC-150 TCSPC Module (Becker & Hickl GmbH, Germany).
  • DCC-100 Module (Becker & Hickl GmbH, Germany).

2.3. Working Components/Optical Filters

Both MFLI and HMFLI platforms mentioned in Subheadings 2.1 and 2.2 make use of optical filters that serve to block the excitation light and only collect the sample’s emissions towards the detector. The optical filters used for the targeted fluorescent dyes and samples are the following (see Note 1):

• FF01-780/12–25 (Semrock, NY).

• FF01-720/13–25 (Semrock, NY).

• FF01-715/LP-25 (Semrock, NY).

2.4. NIR FRET Pairs

• AF700-AF750 NIR FRET pair

• 10 μg/μL AF700 (FRET donor, Thermo Fisher Scientific, A20010).

• 10 μg/μL AF 750 (FRET acceptor, Thermo Fisher Scientific, A20011).

• AF700-QC-1 NIR FRET pair.

• 10 μg/μL AF700 (FRET donor, Thermo Fisher Scientific, A20010).

• 10 μg/μL dark quencher IRDye QC-1 (Li-Cor, 929-70030).

• Fluorophore dyes are dissolved in dimethyl sulfoxide (DMSO) (Sigma D2438).

2.5. Fluorescently Labeled Antibody and Ligand Probes

• 1 mL of 7 mg/mL human iron-bound holo-Tf (Sigma T4132).

• 1 mL of 1 mg/mL humanized monoclonal antibody Trastuzumab (TZM; generously provided by Genentech, Inc).

• 100 μL of 1 M NaHCO_3, pH 8.3.

• Phosphate buffered saline (PBS), pH 7.4.

• Amicon Ultra-4 centrifugal filter units (Sigma Z648035, MWCO 30 kDa).

• Spectrophotometer DU 640 (Beckman Coulter, Fullerton, CA, USA).

• NanoDrop Lite spectrophotometer (Thermo Fisher Scientific, USA).

• Acrodisk syringe filters 0.2 μm (PALL Life Sciences PN 4602).

• Sterile 1 mL tuberculin syringes.

• Custom Tf-QC-1 conjugation can be performed by Li-Cor (Lincoln, NB, USA) with average dye to protein ratio of ~3.

2.6. Breast Tumor Xenograft Models

• Breast cancer human cell line T47D (HTB-133) displaying elevated expression of TfR is obtained from ATCC and cultured in DMEM (Thermo Fisher #11965) media, supplemented with 10% fetal bovine serum (ATCC, 30-2020), 4 mM L-glutamine (Thermo Fisher, 25030081), and 10 mM HEPES, pH 7.5 (Sigma, H3375).

• Breast cancer human cell line AU565 overexpressing HER2 (CRL-2351) is obtained from ATCC and cultured in RPMI 1640 (Thermo Fisher, 22400089) media, supplemented with 10% fetal bovine serum (ATCC, 30-2020).

• Cultrex BME Type 3 (R&D Systems, 3632-005-02).

• Sterile 1 mL tuberculin syringe with sterile 26-gauge needles.

• Athymic CrTac:NCr-Foxn1nu nude 4–5 weeks old female mice (Taconic, NY).

• Chlorophyll-free diet (Altromin, Lage, Germany).

• Digital tumor volume caliper.

• Isoflurane gas anesthesia system (EZ-SA800 System, E-Z Anesthesia Systems, Inc).

• Digitally controlled warming pad (Rodent Warmer X1, Ugo Basile, Gemonio (VA) Italy).

• Sterile lubricant eye ointment.

2.7. Immunohistochemical Validation

• Vectastain Elite ABC HRP kit (Vector Laboratories; PK-6101).

• Vector NovaRED peroxidase substrate (Vector Laboratories; SK-4800).

• 0.5% Methyl Green.

• Antibodies used and their respective manufacturers are listed in Table 1.

Table 1.

Tumor markers used for IHC validation

Description	Marker	Epitope retrieval	Dilution	Source	Catalog number
Validation of NIR human probe tumor binding	Tf	10 mM Na citrate, pH 6	1:2000	Abeam	1223
	TZM	1 mM EDTA, pH8	1:100	R&D Systems	MAB95471-100
Breast tumor tissue marker	Estrogen receptor	10 mM Na citrate, pH 6	1:80	NeoMarkers	RB-1493-PO
	HER2	1 mM EDTA, pH8	1:800	Cell Signaling	2165
	TfR	10 mM Na citrate, pH 6	1:250	Abcam	84036
Glucose metabolism	GLUT1	10 mM Na citrate, pH 6	1:200	Thermo Fisher	PA1-46152
Vascular tissue	CD31	TRIS/EDTA, pH 9	1:50	Thermo Fisher	PA5-16301
Cancer tissue marker	Vimentin	10 mMNa citrate, pH 6	1:200	Thermo Fisher	MA5-16409

3. Methods

3.1. Biological Models and Sample Preparation

In vitro probe preparation and further injections on animal models for in vivo imaging are herein described. The application of anesthesia and temperature regulation as required by animal safety protocols are also explained.

3.1.1. Probe Labeling and Purification

1. Tf and TZM are labeled with 20 μL and 8 μL respectively of 10 μg/μL fluorophores dissolved in DMSO (see Subheading 2.4) in the presence of NaHCO₃ buffer according to manufacturer instructions (see Notes 2 and 3) [45, 49].

2. Probes are washed and purified by using centrifugal filter units (MWCO 30 kDa) centrifuged at 3000 g for 10 min.

3. After four washes with 3–4 mL PBS, the probes are reconstituted in 0.5 mL PBS.

4. Protein concentration is measured by NanoDrop Lite spectrophotometer (1A/cm) and normalized to 1 mg/mL.

5. Degree of labeling, typically ~1.5–2.5 fluorophores per molecule, is determined by spectrophotometer DU 640.

6. Probes are sterilized using syringe filters (0.2 μm) and stored at 4 °C.

3.1.2. Breast Tumor Xenograft Models

1. For breast tumor xenograft models, athymic nude 4–5 weeks old female CrTac:NCr-Foxn1nu mice are orthotopically injected with 10 × 10⁶ cancer cells in PBS mixed 1:1 with Cultrex BME Type 3, into inguinal mammary fat pad (total volume 200 μL).

2. Tumor growth will be monitored daily and the tumor size will be restricted to 500 mm³ as determined by tumor volume caliper measurements.

3. To minimize interference from endogenous auto-fluorescence, mice are given a chlorophyll-free diet (Altromin, Lage, Germany).

4. The tumors are allowed to grow for 3–4 weeks before imaging.

5. Nude mice are kept in cages held on HEPA-filtered ventilated cage racks. Nude mice are immunocompromised and thus should be maintained in specific pathogen-free (SPF) rooms that are carefully monitored for the presence of mouse pathogens.

3.1.3. Tf-TfR In Vivo Imaging Assay

1. To image and quantify Tf-TfR binding and internalization in tumors, i.e. target engagement, human breast cancer cell line T47D is used for tumor xenograft model due to its high expression of TfR.

2. When Tf labeled either with donor (Tf-AF700) or acceptor (Tf-AF750) fluorophore binds homodimeric TfR, FRET occurs, resulting in quenching donor fluorophore lifetime which is quantified as FRET signal (Fig. 1a). TfR–Tf complexes continue to undergo FRET events during the whole process of Tf–TfR complex endocytosis via clathrin-coated pits and subsequent intracellular endosomal recycling [28, 29].

3. These FRET events are registered by MFLI or HMFLI imagers by sensing the reduction of donor’s lifetime (see below) [31, 37, 45, 46].

Fig. 1.

Illustration of FRET events using a Tf-TfR (a) and TZM-HER2 (b) target engagement model systems. The donor- and acceptor-labeled probes bind their respective dimerized target receptor which results in FRET occurrence

3.1.4. TZM-HER2 In Vivo Imaging Assay

1. Unlike Tf-TfR target engagement, which can be detected at various levels in multiple types of tumor xenografts, TZM imaging requires cancer models that overexpress HER2, such as AU565 human breast cancer cell line.

2. Binding therapeutic monoclonal antibody TZM labeled either with donor (TZM-AF700) or acceptor (TZM-AF750) fluorophore to dimeric HER2 results in quenching donor fluorophore lifetime which is quantified as FRET signal (Fig. 1b) [49].

3. These FRET events are registered by MFLI or HMFLI imagers by sensing the reduction of donor’s lifetime (see below) [31, 37, 45, 46].

3.1.5. Mice Injections for In Vivo Imaging

1. Animals are anesthetized using an isoflurane based anesthesia breathing machine.

2. Once anesthetized with isoflurane, nude mice are retro-orbitally injected with donor only (A:D = 0:1; for example, Tf-AF700 or TZM-AF700) or NIR FRET pair labeled probes (A:D = 2:1; for example,Tf-AF700/Tf-AF750 or TZM-AF700/TZM-AF750); increasing A:D ratios can also be used (see Note 3). For donor probe, concentration is kept at 20 μg/mL, whereas acceptor probe is at 40 μg/mL in sterile PBS. FRET pair labeled probes are mixed shortly before injection.

3. Solutions administered to mice are prepared in sterile conditions, such as they are made in sterile microcentrifuge tubes with sterile buffer solutions using sterile pipettes in a sterile environmental hood.

4. TZM imaging typically starts at 24 h post-injection. Tf signal in tumors can be detected as early as 2 h post-injection.

3.1.6. Anesthesia and Temperature Control for In Vivo Imaging

1. In preparation for imaging session, animals are anesthetized using anesthesia breathing machine.

2. After initial anesthesia induction, the flow is switched to the nosecone breather, mouse is placed in a supine position on digitally controlled warming pad at the imager stage set to 39 °C and secured by taping the limbs to the stage.

3. Sterile lubricant is applied to the eyes to prevent drying.

4. The depth of anesthesia is monitored by checking each mouse’s slow breathing rate and lack of the response to a foot pinch.

3.2. Gated-ICCD MFLI of Biological Models

The gated-ICCD MFLI apparatus, set to image in reflectance mode, is comprised of the components listed in Subheading 2.1. The setup, data acquisition, post-processing, and analysis methodology are detailed briefly herein.

3.2.1. Setup for Gated-ICCD MFLI Imaging

1. The Mai Tai laser (tunable Ti:Sapphire) output with power regulated by the CCVA-TL-KT control module is coupled into a 400 μm core-diameter fiber through fiber coupler with objective M-10X.

2. The fiber is then coupled to a collimator and lens to distribute light in the illumination DMD. This one projects the wide-field illumination to the sample plane as illustrated in Fig. 2.

3. Light from the sample plane is collected via the gated-ICCD. Adjusting the DMD to the sample plane is necessary for optimal data quality. One should ensure both homogeneity of illumination and proper focus across the field of view (FOV).

4. The Trigger Delay Unit, preprogrammed to shift the trigger by a fixed increment, outputs a TTL pulse of 5 V after a set time-delay.

5. Upon receiving the TTL pulse, the HRI’s photo-cathode switches potential and opens the gate for a predetermined gate-width (for example, 40 ps commonly used).

6. The ICCD camera integrates an amplified signal over a predetermined exposure time for several laser pulses. Once the exposure time is reached, the ICCD-obtained signal is read-out and recorded as a time-gate.

7. The succeeding time-gate measurement is obtained via shifting of the trigger by a unit delay increment, repeating the process above for n total time-gates, or, until the complete temporal signature made up of Temporal Point Spread Functions (TPSFs) has been acquired.

Fig. 2.

Schematics of the gated-ICCD MFLI system. Suggested system settings for use during in vivo imaging are provided on the boxes at the right

3.2.2. Data Acquisition Procedure

For proper data acquisition:

1. Ensure that the triggering pulses for the illumination (Mai Tai 80 MHz frequency) and Photon Counting related unit are in sync.

2. Light from the sample plane is collected via the gated-ICCD in “DC mode” in order to adjust sample plane illumination and focus to user satisfaction.

3. Ensure fluorescent sample is properly placed and secured on imaging stage. For in vivo small animal imaging applications, both anesthesia and temperature control (such as electronic warming pads) must be placed effectively.

4. Ensure laser excitation wavelength is set accordingly for the application of interest. For FRET-MFLI, the laser’s wavelength should be set close to that of the donor’s excitation maximum. Take note of the laser power’s stability, as some lasers require warm-up and/or some period of wait-time before satisfactory stability is reached.

5. Acquire the Instrument Response Function (IRF) across the FOV of the sample plane. This should be performed without any use of an emission filter.

6. Afterwards, ensure that application-specific filter(s) are placed onto the gated-ICCD. For the case of FRET, both a band-pass and long-pass should be used in order to maximize data signal-to-noise ratio (SNR) and minimize unwanted signal bleed-through [37, 45, 52]. Laser power may need adjustment.

7. Light from the sample plane is collected once more via the gated-ICCD in “DC mode” to maximize the SNR of TPSF acquisition. The multichannel plate gain will likely require modification depending on the application to ensure adequate data quality (i.e. for in vivo applications, 550 V is commonly employed). Further, a modest increase in exposure time and/or widening of the camera’s aperture can improve photon-count.

8. Acquire gated-ICCD MFLI data (see Note 4).

3.2.3. Intensity Image Retrieval

The gated-ICCD apparatus can be set to continuous wave (CW) manually via the HRI. Otherwise, one can obtain the intensity from a MFLI acquisition through post-processing by taking the maximum over the temporal dimension.

3.2.4. Lifetime Image Retrieval

To date, retrieval of lifetime in MFLI has been validated through three main inverse solving techniques:

1. Usage of a Least-Squares-based iterative fitting procedure (Fig. 3) [53]. Iterative fitting is a commonly utilized approach for pixel-wise estimation of lifetime parameters through minimization of the sum of residual error between an exponential model (Γ(t)) and an experimentally acquired fluorescence decay of interest. In the case of FRET quantification, wherein the lifetime of both the quenched and nonquenched donor fluorophore contribute in varying degrees at each pixel, a bi-exponential model is used:

   $$\Gamma \left(t\right)=IRF\left(t\right)\times I\left[A_R{e}^{\raisebox{1ex}{$-t$}\!\left/ \!\raisebox{-1ex}{${\tau }_1$}\right.}+\left(1-A_R\right)e^{\raisebox{1ex}{$-t$}\!\left/ \!\raisebox{-1ex}{${\tau }_2$}\right.}\right]$$

   where I, τ₁ and τ₂, A_R, correspond to the signal intensity, the short lifetime (quenched donor), long lifetime (unquenched donor) and fractional amplitude of the short lifetime, respectively. The value of t is set according to the system settings. Further, for proper retrieval of the A_R the lower and upper bound of both τ₁ and τ₂ are set within known values (commonly utilized windows range 200 ps). The previously mentioned acquisition of the IRF(t) (Subheading 3.2.2) is used for signal deconvolution and the experimental decay is subsequently max-normalized (divided by scalar value I) prior to beginning model-based lifetime estimation (see Fig. 3a) [26, 45, 46].

2. Other validated approaches for MFLI-FRET retrieval are classified as “fit-free,” and include: phasor approach [54] and image reconstruction through deep learning [45]

Fig. 3.

Example of a single pixel’s TPSF, corresponding IRF (a) and bi-exponential parametric fitting via IRF deconvolution followed by a bounded minimization of residual error (b). This process is pixel-wise (mean-lifetime (τ_M) of a single fit indicated spatially by red arrow, (c) and thus iterates over all TPSFs of interest included in the FOV. The well-plate results given (c) illustrate an expected increasing trend in mean-lifetime with decreased A:D ratio (from 3:1–0:1 acceptor to donor concentration where, in this example, 1 represents 50 μg/mL donor concentration)

3.3. Hyperspectral MFLI of Biological Model

The assembled HMFLI system composed of elements listed in Subheading 2.2 assembles to acquire hyperspectral time domain data in reflectance mode. The setup, acquisition of time-domain data, and further inverse solving of the sample’s image are herein described.

3.3.1. Setup for HMFLI Imaging

1. The Mai Tai laser output with power regulated by the CCVA-TL-KT control module is coupled into a 400 μm core-diameter fiber through fiber coupler with objective M-10X. The fiber is then coupled to a collimator and lens to distribute light in the illumination DMD. This one projects the wide-field illumination to the sample plane as illustrated in Fig. 4.

2. The emissions from the sample plane are then detected by a DMD, which adopts a different pattern per measurement. The DMD must be correctly aligned with the area of interest in the sample plane.

3. Structured emissions going out of the DMD travel through a fiber bundle that connects to the spectrophotometer unit (see Note 5).

4. Light in the spectrophotometer is directed through mirrors and passed by the diffraction grating to split and diffract the emission into its different wavelength components.

5. These emissions then travel to the 16-channel PMT detector and a wavelength range is detected per channel (see Note 6).

6. The PMT signals are relayed to the SPC-150 and DCC-100 PCI cards (see Note 7).

7. The SPC-150, in correlation to the triggering signals received from the laser’s OCF trigger generator, forms a TPSF per measurement (DMD structured pattern) (see Note 8).

8. Previously to detecting the HMFLI data, a NIR CCD camera is set to see the same image plane as the detection DMD. This helps to control the position of the sample with respect to the DMD.

Fig. 4.

Schematics of the gated Hyperspectral MFLI system and its main components. Mai Tai Laser and Power Control Module are controlled through a separate computer not shown here. A structured illumination Hadamard pattern is displayed as example. The mirror displayed on the schematic is movable and is placed on top of the image plane only for the acquisition of the “ground-truth” sample intensity image with external NIR ICCD camera

3.3.2. Data Acquisition Procedure

For the data acquisition process:

1. Make sure triggering pulses between the illumination source (Mai Tai 80 MHz frequency) and SPC unit sync. As well verify system alignment.

2. Prepare fluorescent model. In case of in vivo imaging, anesthesia and temperature control systems must be adapted to the imaging platform for usage during the procedure.

3. Set appropriate emission filter in front of the detection DMD. For example, filter FF01–715/LP-25 for detecting emissions of FRET between AF700/AF750 fluorescently labeled probes.

4. Set wide-field illumination depending on application (see Note 9). For example, 695 nm excitation for AF700- and AF750-based FRET excitation. Set a low power and acquisition time. These parameters can be later adjusted depending on the amount of fluorescent signal detected on the PMT.

5. Start acquisition process, where the detection DMD will vary one pattern per measurement. Further information of the pattern formation can be found in [55]. Once the imaging is finished, a 3D measurement matrix of size C × T × P is formed, where C is the number of detection channels, T is the number of time channels, and P the number of recorded patterns.

6. Replace the optical filter with a Neutral Density Filter to attenuate the illumination light. Acquire an extra acquisition to obtain the IRF (see Note 10).

3.3.3. Intensity Image Retrieval

Since the HMFLI platform has a single multichannel PMT an inverse solving procedure must take place for a TPSF to be retrieved per pixel of the desired resolution.

1. Use of minimization algorithm to inverse solve Px(t) = M(t) for the sample plane image x(t), where P represents the acquisition patterns projected on the detection DMD scaled to an NxN resolution and M(t) is the acquired 3D measurement matrix (see Note 11).

2. Usage of Least-Squares-based [56] and TVAL3 [57] minimization have been reported. Furthermore, deep learning has been recently proposed for this function [57, 58].

3. The inverse solved x(t) matrix is of size NxNxT, where values are integrated over the dimension of time-channels (T) forming an NxN intensity image of the sample plane, as shown in Fig. 5.

Fig. 5.

(a) HMFLI output measurements m(t) for a pattern P of NxN resolution. By having both m(t) and P, the time-domain distribution of the sample is retrieved for an NxN resolution. (b) The time domain inverse solved data x(t) contains one TPSF per pixel of the NxN image space. The TPSFs are bi-exponentially fitted for each pixel of x(t) to retrieve lifetime per pixel for lifetime maps. A reconstruction of a FRET well-plate containing different ratios of Tf-AF750 acceptor to Tf-AF700 donor is exemplified. The first well on row one contains Tf AF700 only at a 50 μg/mL concentration, while row two contains varying acceptor to donor concentration ratios of 1:1 to 3:1. The last row contains the acceptor-only probes in the same concentrations as in row two. To exemplify, a lifetime map is displayed for 1 out of 16 channels where the donor signal is the highest

3.3.4. Lifetime Image Retrieval

To retrieve lifetime values the inverse solved matrix x(t) of size NxNxT and which contains one TPSF per pixel of the NxN space is fitted on the fluorescent decay part of the TPSF in a similar procedure to the one mentioned in Subheading 3.2.4 for the Gated MFLI system.

1. Convolve the IRF with the TPSF of each pixel in the NxNxT matrix.

2. Normalize both IRF and TPSFs.

3. Smooth the TPSF using a Gaussian-anscomb filter in case the TPSF has a low signal-to-noise ratio.

4. Use same bi-exponential fitting routine as mentioned in Subheading 3.2.4 for each TPSF on the pixel space to obtain lifetime maps as represented in Fig. 5b.

3.4. Immunohistochemical (IHC) Validation

1. Validation of NIR probe binding to tumor xenografts requires IHC staining using antibodies against those specific human probes (Table 1).

2. Tumor binding can be validated by staining against tumor tissue markers; examples are provided in Table 1.

3. After last imaging session, mice are euthanized, excised tumors are fixed in buffered formalin and paraffin embedded.

4. Consecutive 7 μm-thick sections of tumor tissues are processed for various IHC stainings using standard protocol and reagents from Vectastain Elite ABC HRP kit and Vector NovaRED peroxidase substrate (see Note 12).

5. The tissues are counterstained with 0.5% Methyl Green solution for 1 min followed by washing in water, differentiation in 70% ethanol for 1 min, dehydration in graded ethanol baths, and coverslipping.

4. Notes

1. Optical filters are placed on the detection optical path so that detectors only receive the targeted fluorophores emissions.

2. To avoid self-quenching and obscuring binding site for in vivo FRET imaging, it is important to maintain degree of labeling between 1.5 and 3 fluorophores per protein molecule. If scaling up conjugates, it is recommended to split the reaction solution in two Amicon columns to avoid membrane clogging and probe loosing.

3. Donor only provides a negative FRET control and a sample for measuring the donor unquenched long lifetime component, as described in Subheading 3.2.4. Donor plus acceptor samples provide a FRET sample for measurement of quenched short lifetime component as described in Subheading 3.2.4. Acceptor: Donor ratios at 2:1 are routinely used for TfR-Tf and HER2 target engagement MFLI imaging but increasing A:D ratios can also be used for optimizing MFLI imaging of other potential receptor–ligand systems.

4. To avoid saturation of the detector (and potential damage) it is important to adjust the power of the laser as well as the structured illumination patterns to ensure optimal SNR of the fluorescence data acquired over the whole body of the animal [41, 51, 59].

5. DMDs and optical components are adapted for the NIR range to match the excitation and emission ranges of the probes.

6. Wavelength range can be tuned by choosing a central wavelength through a micrometer screw that operates the grating inside the spectrophotometer. Ranges can vary from 400–1100 nm, with 4.5 nm space in between channels.

7. SPC-150 and DCC-100 PCI cards are inserted on the CPU and receive the trigger signal from the illumination source to help perform the timed photon counting process. The DCC unit assists the detector power regulation.

8. Trigger signals must perfectly match the time intervals specified on the detector unit, otherwise both the units will not be in sync.

9. The focal distance from the image sample plane to the detection DMD must be optimized. In case of nonplanar samples like mice the focal length from the illumination and to the detection units should be adapted depending on the thickness of the mouse.

10. IRFs should optimally be acquired at each spectral channel due to spectrally distinct responses (Full width at half maximum-FWHM- of IRF and time characteristics).

11. The pattern changes per measurement and the full pattern basis can be changed to Fourier, Wavelet or different variations of patterns. In the case of this application, Hadamard patterns organized by spatial frequency have shown improved performance [55, 56].

12. We use by default epitope retrieval by boiling sections in suitable buffer for 20–30 min and cooling down for 20 min.