A humanized antibody for imaging immune checkpoint ligand PD-L1 expression in tumors

Antibodies targeting the PD-1/PD-L1 immune checkpoint lead to tumor regression and improved survival in several cancers. PD-L1 expression in tumors may be predictive of response to checkpoint blockade therapy. Because tissue samples might not always be available to guide therapy, we developed and evaluated a humanized antibody for non-invasive imaging of PD-L1 expression in tumors. Radiolabeled [111In]PD-L1-mAb and near-infrared dye conjugated NIR-PD-L1-mAb imaging agents were developed using the mouse and human cross-reactive PD-L1 antibody MPDL3280A. We tested specificity of [111In]PD-L1-mAb and NIR-PD-L1-mAb in cell lines and in tumors with varying levels of PD-L1 expression. We performed SPECT/CT imaging, biodistribution and blocking studies in NSG mice bearing tumors with constitutive PD-L1 expression (CHO-PDL1) and in controls (CHO). Results were confirmed in triple negative breast cancer (TNBC) (MDAMB231 and SUM149) and non-small cell lung cancer (NSCLC) (H2444 and H1155) xenografts with varying levels of PD-L1 expression. There was specific binding of [111In]PD-L1-mAb and NIR-PD-L1-mAb to tumor cells in vitro, correlating with PD-L1 expression levels. In mice bearing subcutaneous and orthotopic tumors, there was specific and persistent high accumulation of signal intensity in PD-L1 positive tumors (CHO-PDL1, MDAMB231, H2444) but not in controls. These results demonstrate that [111In]PD-L1-mAb and NIR-PD-L1-mAb can detect graded levels of PD-L1 expression in human tumor xenografts in vivo. As a humanized antibody, these findings suggest clinical translation of radiolabeled versions of MPDL3280A for imaging. Specificity of NIR-PD-L1-mAb indicates the potential for optical imaging of PD-L1 expression in tumors in relevant pre-clinical as well as clinical settings.


INTRODUCTION
Tumor cells avoid the immune response by exploiting immune checkpoints through expression of immunosuppressive molecules, recruitment of suppressive immune cell populations and secretion of soluble suppressive factors [1]. Immune checkpoints are inhibitory pathways integral to the immune system, which are critical for modulating the immune response to maintain self-tolerance and prevent autoimmunity. Most immune checkpoints involve ligand-receptor interactions, the inhibition of which repeal the immunosuppression exerted by tumor cells, leading to recognition and destruction of tumor cells by the immune system [1,2]. Treatment with antibodies blocking immune checkpoint pathway ligands and receptors has shown durable tumor regression and improved patient survival [1,[3][4][5][6]. Programmed death ligand-1 (PD-L1, B7-H1 or CD274) is emerging as a www.impactjournals.com/oncotarget central player in immune checkpoint therapies.
PD-L1, a 290 amino acid type I transmembrane glycoprotein, is the primary ligand of programmed death-1 (PD-1). Binding of PD-L1 to PD-1 suppresses T-cell immune activity and restricts tumor cell killing [1,7]. PD-L1 expression is upregulated in the tumor microenvironment (TME), (possibly as an immuneevasion mechanism) [8] and may be due to: (i) increased PD-L1 expression on tumor cells by intrinsic oncogenic events (e.g., loss of phosphatase and tensin homolog) [9], (ii) tumor cell PD-L1 induction in response to T-cell secreted interferon-gamma [10] and (iii) PD-L1 expression on accumulated myeloid cells and/or dendritic cells that have suppressive effects on T-cells [11]. In cancer patients, there is a strong correlation between PD-L1 expression on tumor cells and poor prognosis [7]. Furthermore, in PD-1 and PD-L1 targeted therapies across multiple cancer types, there is a strong positive correlation between pre-treatment PD-L1 expression in TME and therapeutic response to PD-1/PD-L1 pathway inhibitions [6,12,13].
Antibodies targeting PD-L1 (MPDL3280A, MEDI4736 and BMS-936559) have demonstrated antitumor activity in diverse tumor types, including renal cell carcinoma (RCC) [14], advanced melanoma [6], non-small cell lung cancer (NSCLC) [6,14,15], and bladder cancer [13] among others [2,16]. Nearly 45% of patients with PD-L1 positive TME show an objective response (OR) following immune checkpoint blockade. Those observations suggest that expression of PD-L1 in tissue biopsies is a valuable biomarker for immune checkpoint therapies [2,16,17], but tissue samples are often impractical to obtain, particularly in the setting of recurrent and metastatic disease. Thus, non-invasive detection of the changes in PD-L1 expression in the TME may guide patient management.
The clinical efficacy of MPDL3280A prompted us to investigate its application for non-invasive detection of PD-L1 expression in tumors. We developed radiolabeled and near-infrared (NIR) dye-tagged analogs of MPDL3280A (PD-L1-mAb) and tested their specificity in Chinese hamster ovary tumors with constitutive PD-L1 expression. Observations from those studies were then validated in orthotopic and subcutaneous TNBC and NSCLC human tumor xenografts with varying levels of expression of PD-L1. Specific accumulation of SPECT and optical imaging signal intensity was seen in tumors with high PD-L1 expression. Specificity of the uptake was confirmed by ex vivo biodistribution and immunohistochemistry studies. Our results with a humanized PD-L1-mAb confirm PD-L1-specific signal accumulation in multiple tumor models. These studies demonstrate imaging of PD-L1 in vivo and provide the basis for clinical translation of derivatives of anti-PD-L1 antibodies for imaging.
Blocking of [ 111 In]PD-L1-mAb binding to CHO-PDL1 cells, by addition of 10-fold molar equivalent excess of unlabeled antibody, reduced radioactivity uptake by 75%, indicating that [ 111 In]PD-L1 mAb binding is specific. A similar uptake profile was observed with NIR-PD-L1-mAb ( Figure 1D). These in vitro data indicate that PD-L1 targeted antibody-based imaging probes can be used to detect graded levels of PD-L1 expression in cancer cells.

PD-L1 mAb shows specific uptake in tumors with stable PD-L1 expression
Several factors, as discussed in the introduction, contribute to changes in PD-L1 expression in the tumor microenvironment. Accordingly, we first established the in vivo specificity of [ 111 In]PD-L1-mAb in tumors with constitutive PD-L1 expression. SPECT/CT images acquired over 120 h demonstrated substantial and specific accumulation of [ 111 In]PD-L1-mAb in CHO-PDL1 tumors but not in control CHO tumors (Figure 2A). Radioactivity accumulation could also be seen in the lungs, liver, and spleen.
To validate the imaging results and to establish protein dose requirements, a protein-dose escalation biodistribution study was performed. In mice injected with [ 111 In]PD-L1-mAb alone, at 48 h, the highest uptake (in %ID/g) was in spleen (23.5±8.2), followed by CHO-PDL1 tumor (4.7±0.7) and liver (8.2±4.5) ( Table 1). In contrast, in mice co-injected with 10, 30 and 90 μg of unlabeled antibody, the highest uptake was in CHO-PDL1 tumors, in which %ID/g was 13±4 12.5±0.6, and 13.3±1.6 for 10, 30 and 90 μg dose cohorts, respectively. In the spleen, uptake significantly decreased with increased antibody dose, suggesting that spleen acts as a sink for this specific antibody. There were no significant changes in other tissues, nor in tumor or tissue uptake between 30 and 90 μg dose cohorts. Based on the high CHO-PDL1 tumor-tomuscle (21.7±1.3) and CHO-PDL1 tumor-to-blood ratios (2.5±0.1), all other biodistribution studies were performed with 30 μg co-injection of unlabeled antibody.
We further evaluated the temporal changes in [ 111 In] PD-L1-mAb (30 μg dose) biodistribution in the CHO-PDL1 tumor model at 120 h (Table 1). There was a substantial accumulation of radioactivity in CHO-PDL1 tumors (16.5±3.0 %ID/g). Blood pool radioactivity was reduced and spleen uptake was increased at 120h. In all other tissues, there was no significant difference in %ID/g between 48 and 120 h time points. Thus for CHO-PDL1 tumors, by 120 h, there was high tumor-to-muscle (38.7±8.0) and tumor-to-blood ratios (5.8±1.9), accounting for the tumor-specific high image contrast seen in the SPECT/CT images. In vivo PD-L1 specificity of the antibody was further validated in mice that received 1.5 mg of unlabeled PD-L1 mAb as blocking dose, in which there was a 65% decrease (P < 0.001) in CHO-PDL1 tumor radioactivity uptake (Table 1). That tumor specific radioactivity uptake was corroborated by intense immunoreactivity observed in CHO-PDL1 tumors (Supplementary Figure 3).
Similarly, in NIR-PD-L1-mAb-injected mice, signal intensity was consistent and substantial in CHO-PDL1 tumors during the 120 h study ( Figure 2B). Biodistribution studies at 120 h showed a 3-fold increase in fluorescence signal intensity in the CHO-PDL1 tumors compared to CHO control tumors (P < 0.0001, Figure 2C and 2D). Liver, lungs, and kidneys also showed a distinct accumulation of fluorescence signal. These results show that NIR-PD-L1-mAb can be used to specifically detect tumor PD-L1 expression in vivo.

Increased uptake of PD-L1 mAb in TNBC xenografts with high PD-L1 expression
SPECT/CT images of mice with MDAMB231 and SUM149 TNBC xenografts that were injected with [ 111 In] PD-L1-mAb showed high radioactivity accumulation in the MDAMB231 tumors compared to SUM149 ( Figure  3A). The distribution profile in other tissues was similar to that of the CHO tumor model. Biodistribution studies indicated highest radioactivity in MDAMB231, blood and spleen ( Figure 3B). The %ID/g values for the MDAMB231 and SUM149 tumors were 8.9±0.26 and 5.5±0.21, respectively at 72 h after [ 111 In]PD-L1-mAb injection. The tumor-to-muscle and tumor-to-blood ratios for MDAMB231 tumors at 120 h were 8.2±0.8, and 0.79±0.06, respectively. IHC evaluation showed strong PD-L1 immunoreactivity in MDAMB231 but not in SUM149 tumors ( Figure 3C), validating [ 111 In]PD-L1-mAb specificity.
Optical imaging of NIR-PD-L1-mAb distribution in the MDAMB231 and SUM149 xenografts during 120 h showed consistent and high fluorescence intensity in the MDAMB231 tumors compared to the SUM149 tumors  Figure 3D). Ex vivo analysis of the same mice confirmed the imaging observations ( Figure 3E). In addition to the MDAMB231 tumors, increased fluorescence intensity was observed in SUM149 tumors, liver and lungs ( Figure   3E and 3F). These studies establish that [ 111 In]PD-L1-mAb and NIR-PD-L1-mAb have the specificity to detect endogenous PD-L1 expression in TNBC tumors. Increased uptake of PD-L1 mAb in NSCLC xenografts with high PD-L1 expression SPECT/CT imaging of mice showed high accumulation of radioactivity in subcutaneous H2444 tumors by 120 h, compared to H1155 tumors ( Figure  4A), and the %ID/g for H2444 and H1155 tumors at 144 h were 7.46±0.12 and 3.63±0.57, respectively ( Figure  4B). The tumor-to-muscle and tumor-to-blood ratios for H2444 tumors were 8.9±0.8 and 0.9±0.2, respectively. IHC evaluation showed strong PD-L1 immunoreactivity in H2444 tumors but not in H1155 tumors ( Figure  4C), validating [ 111 In]PD-L1-mAb specificity. In mice with orthotopic H2444 lung tumors, there was specific radioactivity accumulation delineating the tumors from normal lungs by 72 h, and significant enhancement in contrast at 120h ( Figure 4D).
Optical imaging of H2444 and H1155 tumor bearing mice injected with NIR-PD-L1-mAb showed   Figure 4E). This was confirmed by ex vivo biodistribution analysis, which showed a nearly two-fold increase in signal intensity in the H2444 tumors compared to H1155 ( Figure 4F and Supplementary Figure 4). High fluorescence intensity was also observed in liver and kidneys. Collectively, these studies establish the feasibility of imaging endogenous PD-L1 expression in NSCLC tumors.

DISCUSSION
Preclinical evaluation of a humanized radiolabeled anti-PD-L1 antibody, [ 111 In]PD-L1-mAb, shows specific and increased uptake of radioligand in CHO tumors with stable PD-L1 expression compared to control CHO tumors. The in vivo specificity of [ 111 In]PD-L1-mAb was confirmed by differential uptake in human breast and lung tumor xenografts with endogenous high and low PD-L1 expression, and by studies with the fluorophore-conjugated antibody, NIR-PD-L1-mAb. The results demonstrate a new and non-invasive means to detect PD-L1 expression in tumors with an antibody ready for clinical translation.
We first characterized the distribution of [ 111 In] PD-L1-mAb and protein dose effect on antibody biodistribution in CHO tumors with constitutive PD-L1 expression [19]. Imaging and biodistribution studies showed that [ 111 In]PD-L1-mAb uptake in the tumors was PD-L1 specific and that overall tissue distribution and tumor uptake were concentration dependent. Co-injection of unlabeled antibody prolonged circulation of [ 111 In]PD-L1-mAb, significantly increased the tumor uptake and reduced spleen uptake. These studies suggest that predosing with unlabeled antibody will improve tumor uptake of [ 111 In]PD-L1-mAb and confirm reports from other PD-L1 antibodies [19,20]. This could be possible because high doses of PD-L1 antibody could saturate the PD-L1 expression in splenocytes, increasing availability of the radioactive antibody to bind target sites within tumor. The high tumor-to-muscle we observed at 72 and 120 h, suggest that this would be the optimal time to image tumors with the [ 111 In]PD-L1-mAb.
We then tested the specificity of the [ 111 In]PD-L1-mAb in TNBC xenografts with endogenous PD-L1 expression. TNBC has one of the poorest survival rates [21]. In patients with metastatic TNBC, MPDL3280A has shown promising clinical activity, providing a new therapeutic opportunity for a cancer subtype that heavily relies on chemotherapy [2]. Better understanding of the TNBC response to immunotherapy could be improved by PD-L1 imaging. Our studies with [ 111 In]PD-L1-mAb demonstrated specific uptake in MDAMB231 tumors with high PD-L1 expression, compared to that in SUM149 tumors that have low PD-L1 expression, suggesting that non-invasive PD-L1 detection is a viable option for TNBC. Similar results were also observed in other studies [19]. The values we report for tumor %ID/g in the MDAMB231 tumors are lower than those recently reported by Heskamp et al. (2015) [19]. In addition, our flow cytometry analysis showed that approximately 30% of MDAMB231 cells were positive for PD-L1 expression, which was markedly lower than the > 90% of positive cells reported by Heskamp et al. [19], accounting for the differences in %ID/g. Nevertheless, enhanced signal intensity accumulation and delineation of PD-L1 positive breast tumors is evident in our studies.
Immune checkpoint-targeted therapies have resulted in prolonged tumor regression and improved overall response rates, not only in immunogenic cancers such as melanoma and RCC, but also in cancers not believed to be immunogenic such as lung cancers. Treatment with MPDL3280A results in durable responses in NSCLC patients [6], and non-invasive PD-L1 detection in NSCLC may improve patient stratification. In an orthotopic lung tumor model (H2444), mice injected with [ 111 In]PD-L1-mAb demonstrated SPECT signal in tumor regions that were spatially discrete from the normal lung, and aligned with tumor masses seen on CT by 72 h. Similarly, subcutaneous NSCLC xenografts showed specific uptake in H2444 tumors compared to H1155 tumors. The %ID/g values for H1155 were similar to CHO and SUM149 xenografts with low PD-L1 expression levels. However, %ID/g value for the H2444 tumor was found to be lower than what was anticipated based on the MFI values. Several factors can influence antibody uptake [22] and need to be investigated, including rapid tumor growth, enhanced permeability and retention, and tumor vascularity. Poor antibody delivery to the tumors could reduce therapeutic efficacy, despite their positive PD-L1 status. Non-invasive imaging, as we have demonstrated, could identify the disagreement between PD-L1 status and antibody delivery to the tumor. The clinical utility of such antibody imaging agents would be to use the radiolabeled antibody accumulation in the tumors to guide therapeutic antibody dosing and correlate that uptake with tumor response. This could be used to establish a relationship between tumor PD-L1 status and therapeutic response, which may have prognostic implications.
NIR intra-operative optical imaging has been used for staging and detection of tumor margins during cytoreductive surgery in ovarian cancer [23]. It is also becoming increasingly used to enhance tumor resection in brain surgery (using 5-aminolevulinic acid, scorpion venom, etc.) [24,25], and new agents are emerging specific to a variety of other cancers, including prostate and breast [26][27][28]. With the NIR-PD-L1-mAb, PD-L1 positive tumor contrast was high in PD-L1 positive CHO-PDL1, MDAMB231 and H2444 tumors compared to CHO, SUM149 and H1155 tumors. That specificity suggests that NIR-PD-L1-mAb could be used for noninvasive PD-L1 detection by optical imaging.
Molecular imaging based "optical biopsies" are used for identification of lung nodules in a shorter period of time than standard IHC, and resection of those nodules with high success rates [29]. Similarly, NIR dye labeled EGFR antibody Cetuximab was used to visualize EGFR expression in head and neck squamous cell carcinomas [30]. NSCLCs show significant response rates for immune checkpoint targeted therapies and could benefit from availability of such technologies. Bronchoscopic imaging is routinely used for lung tumor diagnosis and staging but has not been used for PD-L1 expression status in lung tumors [31]. The specificity observed in the lung tumor models we tested indicates that NIR-PD-L1-mAb has the potential for bronchoscopic or thorascopic optical imaging of PD-L1 expression in lung tumors. Imaging PD-L1 expression has recently been demonstrated using a mouse anti-human PD-L1 and hamster anti-mouse antibodies, and an engineered PD-1 derived fragment [19,20,32]. While those studies show the feasibility, the species from which the antibodies were derived and the near certainty of immunogenicity may curtail clinical translation, an experience observed with ProstaScint [33]. We chose MPDL3280A because it is in clinical trials and shows cross-reactivity to both human and mouse PD-L1 [13,18]. Similarly crossreactive Cetuximab was used to image tumor EGFR expression in patients [34]. Demonstrating the power of cross-reactive antibodies to evaluate tumor biology, Cetuximab was also used to detect colon adenocarcinoma in the setting of colitis in immunocompetent mouse models [35]. Such cross-reactivity is more desirable in the case of PD-L1 imaging agents because of the need for immunocompetent model systems to evaluate the immune responses. Importantly, results from preclinical studies using those cross-reactive antibodies, such as the one we have demonstrated using MPDL3280A, may guide clinical studies.
A variety of normal tissues express PD-L1 transcripts, including placenta, lung, liver, spleen, lymph nodes and thymus [7]. [ 111 In]PD-L1-mAb cross-reactivity is reflected in the high radioactivity uptake observed in mouse spleen, lungs and liver in our present study. This is similar to the observations made by Josefsson et al. using an anti-mouse PD-L1 antibody [20]. Antibody crossreactivity may also explain the significant differences in biodistribution reported in our study and the previous study using a mouse anti-human antibody by Heskamp et al. (2015) [19]. The consistently high radioactivity uptake observed in brown fat in our study, known to have immune cells, needs to be further investigated [36]. Because we used immunocompromised mouse models, the radioactivity uptake observed in immune related tissues could be less than one may see in immunocompetent mice. Nevertheless, our results provide an assessment of PD-L1 mAb distribution that could not be observed using either human-only or mouse-only reactive PD-L1 antibodies. Our data show human PD-L1 specific uptake of the antibody in the tumors, and mouse PD-L1 specific uptake in other tissues, and this provides a more detailed perspective on imaging PD-L1.
In summary, we have developed and evaluated nuclear and optical imaging agents for PD-L1 based on a humanized antibody. These represent two new tools that are ready for translation to patients undergoing immune checkpoint therapy. We have demonstrated PD-L1 specific accumulation of nuclear and fluorescent imaging agents in tumors with constitutive PD-L1 expression, and in TNBC and NSCLC xenografts with graded endogenous PD-L1 expression. The results demonstrate the feasibility of noninvasive PD-L1 imaging in vivo. PD-L1 has become an important target for cancer immunotherapy, and PD-L1targeted antibodies have proved effective [6,13]. Based on a humanized antibody that has shown therapeutic promise in patients, the presented data holds considerable potential for clinical translation. The patient management in advanced melanoma, breast and bladder cancers, in which PD-L1 antibodies have shown therapeutic efficacy but lack better response prediction and monitoring strategies, may particularly benefit from non-invasive PD-L1 detection at all the tumor sites.

Cell lines
Four cell lines NCI-H2444 (NSCLC, PD-L1 high ), NCI-H1155 (NSCLC, PD-L1 low ), MDAMB231 (TNBC, PD-L1 high ) and CHO-K1, (henceforth referred to as H2444, H1155, MDAMB231 and CHO respectively), were purchased from the American Type Culture Collection (ATCC) and passaged for fewer than 3 months after which new cultures were initiated from vials of frozen cells. The SUM149 (TNBC, PD-L1 low ) cell line was kindly provided by Dr. Stephen P. Ethier, Medical University of South Carolina, and authenticated by STR profiling at the Johns Hopkins genetic resources facility. SUM149 cells were maintained in Ham's F-12 medium with 5% FBS, 1% P/S and 5 µg/mL insulin, and 0.5 µg/mL hydrocortisone. All other cell lines were cultured in ATCC recommended media in an incubator at 37 o C in an atmosphere containing 5% CO 2 . The CHO-PDL1 cell line generation and maintenance are described in Supplementary Methods.

Flow cytometry
Cells in suspension were harvested by centrifugation with adherent cells detached using enzyme-free, PBSbased cell dissociation buffer (Gibco). The harvested cells were washed twice with flow cytometry buffer (1xPBS with 2 mM EDTA and 0.5% FBS). Cells were stained with anti-human PD-L1 antibody conjugated with phycoerythrin (PE) (catalog #557924, Becton Dickinson) according to the manufacturer's protocol and were analyzed on a FACSCalibur flow cytometer (Becton Dickinson). At least 20,000 events were recorded and analyzed using FlowJo software (Tree Star).
Orthotopic mouse models of lung tumors were generated by injecting lung cancer cells into the left lung of NSG mice. A 1 cm skin incision was made on the left scapula of the anesthetized mouse and the thoracic muscles were separated to expose the costal layer. The H2444 cells, (1×10 6 in 30 µL HBSS containing 50% matrigel), were injected directly though the intercostal space into the left lung, by using a 29G needle on a 0.5 mL syringe. Skin incisions were closed by sutures and antibiotics were applied topically. Growth of the orthotopic lung tumor was monitored by CT imaging.

Preparation of radiolabeled antibody
The

In vitro binding assay and immunoreactive fraction (IF) determination
In vitro binding of [ 111 In]PD-L1-mAb to CHO-PDL1, CHO, MDAMB231, SUM149, H2444, and H1155 cells was determined by incubating 1 μCi of [ 111 In]PD-L1-mAb with 1×10 6 cells (in triplicate for each cell line) for 1h at 37 o C. PD-L1 blocking was performed by adding a 10-fold molar equivalent excess of the non-labeled mAb. After incubation, cells were washed three times with cold PBS prior to counting on an automated gamma counter (1282 Compugamma CS, Pharmacia/LKBNuclear, Inc., Gaithersburg, MD).
To determine the immunoreactive fraction (IF), binding of [ 111 In]PD-L1-mAb to 4, 2, 1, 0.5 and 0.25 million CHO-PDL1 cells was carried out as described above. IF was calculated using the Lindmo assay [37]. All in vitro studies were performed in triplicate and repeated three times.

SPECT/CT imaging and analysis
Whole-body SPECT-CT images were acquired on an X-SPECT small animal SPECT/CT system (Gamma Medica Ideas, Northridge, CA, USA) as described previously [38]. Briefly, after an intravenous injection of approximately 400 µCi of [ 111 In]PD-L1-mAb (100 µg of antibody, n = 3), images were acquired at the specified time points in mice. The tomographic data were acquired in 64 projections over 360 o , at 45 s per projection, using medium energy pinhole collimators. CT images were acquired in 512 projections to allow anatomic coregistration. Images were reconstructed using the ordered subsets-expectation maximization algorithm, and 3D volume rendered decay corrected images were generated using Amira 5.5.0 software (Visage Imaging Inc.). Biodistribution studies were also carried out to confirm the imaging study results in CHO/CHO-PDL1, SUM149/MDAMB231 and H1155/H2444 xenograft models with low or high expression of PD-L1 respectively, at various time points after [ 111 In]PD-L1-mAb injection. The percentage of injected dose per gram of tissue (%ID/g) values were calculated based on signal decay correction and normalization to external 111 In standards, which were measured in triplicate. Biodistribution data shown is mean ± the standard error of the mean (SEM).

Optical imaging
PD-L1 expression in different tumor models was assessed by optical imaging using NIR-PD-L1. The NIR-PD-L1 (22 μg) was injected into the tail vein of the mice (n = 3-5) bearing tumors with low and high expression of PD-L1. Mice were anaesthetized with isoflurane and serial images of the dorsal, left lateral, ventral and right lateral surfaces were captured using the Pearl Impulse Imager in white light and 800 nm channels (Software v2.0, LI-COR Biosciences) at 24, 48, 72, 96 and 120 h post injection. On day 5 after the injection of NIR-PD-L1-mAb, mice were euthanized and tumors and selected tissues were dissected and imaged ex vivo. To quantify the signal, equal sized regions of interest (ROIs) were drawn on tumors and tissues and on an area outside the mouse and representative of background. Mean signal intensity in each ROI was normalized by subtracting the background signal, and used for statistical analysis. Data shown is mean fluorescence intensity values ± SEM.

Immunohistochemistry
Tumor sections were evaluated for PD-L1 expression by immunohistochemistry (IHC). Harvested tumors were fixed in 10% neutral buffered formalin, embedded in paraffin, and 4 μm thick sections were obtained on slides. After deparaffinizing with xylene and alcohol gradients, antigen retrieval was done using 10 mM citrate buffer, pH 6.0 (#S1699, Dako target retrieval solution). Tumor sections were then treated with 3% H 2 O 2 for 10 minutes, blocked with 5% goat serum for 1 h, and then incubated with a primary anti-human PD-L1 antibody (#13684, Cell Signaling) at 1:500 dilution at 4 o C overnight. Subsequently, using Dako CSAII Biotinfree Tyramide Signal Amplification System kit, slides were incubated with secondary antibody, amplification reagent, and with anti-fluorecein-HRP. Finally, staining was carried out by adding DAB chromogen. Sections were counterstained with hematoxylin, followed by dehydration with alcohol gradients, xylene washes and mounted with a cover slip.

Data analysis
Statistical analysis of in vitro receptor binding assay data and ex vivo biodistribution data were performed with Graphpad Prism 6 software using an unpaired two-tailed t test. When P < 0.05, the difference between the compared groups was considered to be statistically significant.