Main

T cell-enhancing immune checkpoint inhibitors (ICIs) have gained their place in cancer treatment with impressive, durable antitumur efficacy in a remarkable variety of tumor types1,2,3. However, response rates vary, and only a subset of patients benefits. A combination with another ICI or other medicines can improve response rates but can also increase the risk of adverse events (AEs)1. This highlights the clinical need for tools to optimize treatment strategies for individual patients. Several biomarkers have been identified to select patients for ICI3. These include programmed death-ligand 1 (PD-L1) expression, tumor mutational burden, deficiency of mismatch repair (dMMR) proteins and a T cell-inflamed gene expression profile4,5,6. However, no single biomarker or combination of biomarkers accurately predicts response to ICI.

CD8+ T cells play an essential role in tumor cell destruction by the immune system. Their presence in the tumor is associated with responses to ICIs across several tumor types6,7,8,9,10. An ICI treatment-emergent increase in CD8+ T cell density in tumor biopsy samples has also been associated with tumor response. Most data are available for patients with advanced melanoma with biopsy samples obtained at different time points following the start of ICIs. For example, increased CD8+ cell density in 25 paired tumor biopsy samples collected after 20–120 days pembrolizumab treatment was associated with response11. Others reported a CD8+ T cell expansion in 13 biopsy samples two weeks after anti-programmed cell death (PD-1) antibody therapy initiation, but this was not the case in a study analysing ten mostly late on-treatment biopsy samples after 0.7–26 months9,10. Sampling bias may influence these differences and considerable heterogeneity can exist within or between different lesions within one patient12,13.

Due to these inherent limitations for invasive tumor biopsies, remarkably little is known about the systemic kinetics and heterogeneity of CD8+ T cell distribution among tumor types and individual tumor lesions in patients. To address this issue, we developed the zirconium-89-labeled one-armed antibody 89ZED88082A targeting CD8a, as antibodies or antibody fragments labeled with zirconium-89 (89Zr) allow noninvasive whole-body visualization of a target with positron emission tomography (PET)14,15,16. First, 89ZED88082A uptake with PET was shown in human CD8-expressing tumors xenografted in mice17. We then performed 89ZED88082A PET scanning in patients with solid tumors before and ~30 days after starting ICI treatment with PD-L1 antibody, or PD-1 antibody with or without CTLA-4 antibody. The primary objectives of the study were to characterize the safety, imaging dose and time points, pharmacokinetics and immunogenicity of 89ZED88082A in patients with solid tumors. Secondary objectives included the potential to image whole-body CD8+ T cells, correlations of CD8 PET imaging data with tumor-based assessments and correlations with clinical outcomes and AE to ICI treatment.

Results

Trial population and safety

Between February 2019 and November 2020, 39 patients were enrolled (NCT04029181). One patient with tracer extravasation was excluded from PET analyses (Table 1). Twenty-two of the 29 consecutive patients included for repeated imaging did undergo this, with a median of 30 days following initiation of ICI treatment (IQI 28–36 days). Seven were not scanned during ICI therapy, because of withdrawal before (n = 1) and during (n = 4) treatment due to disease progression, patient anxiety (n = 1) and COVID-19 restrictions (n = 1).

Table 1 Characteristics at study entry of all evaluable patients

No 89ZED88082A-related side effects occurred. AEs due to ICI were consistent with reports from previous studies (Extended Data Table 1).

In part A, two anti-CD8 tracer protein doses (89ZED88082A + unlabeled, desferrioxamine (DFO)-conjugated one-armed antibody CED88004S) were evaluated: 4 mg (n = 3) or 10 mg (n = 6) with serial PET scans 0 (1 h), 2, 4 and 7 (±1) days after administration, followed by a biopsy of a tumor lesion. The 10 mg dose allowed for sufficient blood pool tracer availability (average day 2 mean standard uptake value (SUVmean) 2.9 (±1.0), day 4 SUVmean 1.9 (±0.3)). Compared to 4 mg, the 10 mg dose showed less and stable splenic uptake, indicating abatement of splenic tracer sink effect (Extended Data Fig. 1a). The 10 mg protein dose visualized tumor lesions and lymphoid tissues (Fig. 1 and Supplementary Video), with highest uptake on days 2 and 4 (Extended Data Fig. 2). In vitro, human peripheral blood mononuclear cells did not internalize the tracer (Extended Data Fig. 3), consistent with PET imaging data showing no further increase in tissue signal between days 2–7. Therefore, in part B, the 10 mg protein dose with PET scanning on day 2 was considered optimal.

Fig. 1: Normal tissue biodistribution of 89ZED88082A.
figure 1

a, Representative 89ZED88082A PET scan maximum intensity projection day 2. A whole-body visualization is available as Supplementary Video. b–e, Axial views of the same scan fused with low-dose CT. Arrows indicate uptake in Waldeyer’s ring, cervical lymph nodes (b), spleen, bone marrow (c), renal cortex, small intestine (d) and inguinal lymph nodes (e). f,g Pretreatment uptake with 95% confidence bands across tissues adjusted for protein dose, projected at 10 mg dose (n = 9), days 0 (1 h), 2, 4 and 7 (±1 day), with mean SUVmean (f) and mean SUVmax (g) for lymph nodes and tonsils, not visible on day 0.

Uptake in tumor lesions at baseline

Baseline 89ZED88082A uptake in all nonirradiated lesions (n = 266 in 38 patients) showed an overall geometric mean SUVmax of 5.6 (geometric coefficient of variation 0.72) on day 2. Lesions were detected in all major organs. Median geometric mean SUVmax per patient was 5.2 (IQI 4.0–7.4). Heterogeneity in tumor uptake was observed between and within patients (intraclass correlation coefficient 0.46; Fig. 2a,b(ii) and Extended Data Fig. 4). In 10 patients, 4 with dMMR tumors, 16 lesions (6 dMMR) showed a pronounced tumor-rim uptake (Fig. 2b and Extended Data Fig. 4f–h). Among the 13 evaluable lesions out of these 16, only 3 had computed tomography (CT) evidence of central necrosis.

Fig. 2: 89ZED88082A uptake in nonirradiated tumor lesions.
figure 2

a, Pretreatment uptake in 266 lesions day 2 after tracer injection, ordered by increasing geometric mean SUVmax per patient, visualizing lesion size and site, and aorta background uptake. , diameter. μ, mean. b, Axial views PET/CT scans, arrows indicate lesions. (i) High, heterogeneous uptake in dMMR duodenal tumor. (ii) Uptake in a triple-negative right breast cancer lesion, moderate uptake in pleural and no to minor uptake in lung lesions. (iii) Minor uptake in perivesical dMMR urothelial cell cancer lesion pretreatment increased with rim pattern during treatment (iv). c, Violin plot SUVmax in lesions (n = 212) per site (lymph nodes n = 99, liver n = 35, bone n = 17, lung n = 42, skin n = 19). d, Violin plot of SUVmax in patients with pMMR (n = 25) and dMMR tumors (n = 9). e, Violin plot of SUVmax in lesions with desert (n = 15) and nondesert (n = 19) immune phenotype before and during treatment in 24 patients. c–e, Violin plots with bottom and top 1% of SUVmax values truncated (c and d, not for e); colored dots are the geometric means per patient (d) or lesion (e); black vertical lines are geometric mean SUVmax 95% CI; white dots within black lines and values below the violin plot the actual geometric means. Two-sided nominal P values were derived from linear mixed models taking clustering within patients (and, if applicable, lesions) into account, using a Wald test under restricted maximum likelihood for three of higher-level factors (c) or a likelihood ratio test under maximum likelihood for two-level factors (d,e). SqCC, squamous cell carcinoma; OCCC, ovarian clear cell carcinoma; HCC, hepatocellular carcinoma; UP, unknown primary.

89ZED88082A uptake was related to the lesion’s organ location and highest in malignant lymph nodes (Fig. 2c). Malignant lymph nodes also exhibited 62% higher SUVmax than normal lymph nodes (95% confidence interval (CI) 45–80%, P ≤0.001). We took two approaches to verify whether potential differences in CD8 tracer uptake did reflect CD8-related tumor characteristics. First, we showed that 89ZED88082A tumor uptake was higher in the 9 patients with dMMR than the 25 with mismatch repair proficient (pMMR) tumors (Fig. 2d). Second, we studied with CD8 immunohistochemistry (IHC) the tumors of 24 patients with 22 pre- and 12 on-treatment samples. This showed four inflamed, 15 stromal and 15 desert phenotypes (Extended Data Fig. 5a and Supplementary Fig. 1). The SUVmax was higher in inflamed or stromal phenotype lesions than desert phenotype lesions before and during treatment (Fig. 2e and Extended Data Fig. 5f). Lesions with a CD8 desert phenotype had a geometric mean SUVmax of 4.3 (95% CI 3.1–6.0), while lesions with a stromal or inflamed phenotype had a geometric mean SUVmax of 7.1 (95% CI 5.4–9.4) (P = 0.018); when presented as a to the physiological muscle background uptake, this difference was not significant (Extended Data Fig. 5e). Localized CD8+ T cell density by IHC correlated with the autoradiography signal magnitude in tumor tissues (τ = 0.45, P = 0.015) (Fig. 3a and Extended Data Fig. 5b–d).

Fig. 3: 89ZED88082A in tumor tissues related to CD8 by IHC.
figure 3

a, Autoradiography image of 89ZED88082A uptake in a dMMR colorectal cancer liver metastasis and accompanying CD8 IHC staining. Areas 1, 3 and 5 with moderate to high CD8 expression; 2 and 4 without CD8 expression. The representative image is shown with evident correlation between IHC CD8 expression and autoradiography signal (n = 16). b, Overview of SUVmax and CD8 IHC expression pattern (density score) in lesions with corresponding paired biopsy samples before and during treatment in ten patients. On the x axis, primary tumor type and location of biopsy are shown. The symbol above the bar indicates the radiographic response of the lesion at six weeks. LN, lymph node.

Source data

As of 13 October 2021, median patient follow-up was 5.6 months; 35 of 38 patients were evaluable for best overall response, 4 patients experienced a complete response (CR), 8 a partial response (PR), 4 stable disease (SD) and 19 progressive disease (PD). Baseline tracer tumor uptake showed a positive trend with best overall response evaluation criteria in solid tumours (RECIST) response (Ptrend = 0.064, Extended Data Fig. 6a), and uptake was 40% (95% CI 0–94%) higher in patients with SD/PR/CR as best overall response during ICI (P = 0.040; Extended Data Fig. 6b,c). Patients with an above-median baseline 89ZED88082A-uptake geometric mean SUVmax (that is, >5.2) showed a trend towards superior progression-free survival (PFS) (median 1.5, 95% CI 1.3 to not reached; versus 3.9, 95% CI 2.6 to not reached, P = 0.058) and had superior overall survival (OS) to patients with an uptake below the median (median 6.5, 95% CI 3.3 to not reached, versus 13.8, 95% CI 11.3 to not reached, P = 0.030) (Fig. 4). Analyzed continuously, baseline 89ZED88082A-uptake geometric mean SUVmax (per standard deviation decrease) showed for PFS a hazard ratio (HR) of 1.60 (95% CI 1.03–2.78; P = 0.034) and for OS that of 1.59 (95% CI 1.04–2.72; P = 0.031).

Fig. 4: 89ZED88082A uptake related to tumor response.
figure 4

a,b, PFS (a) and OS (b), according to baseline geometric mean SUVmax below and above median, and with two-sided nominal P values derived from a log-rank test. c, Changes during repeated imaging in tumor uptake and anatomic size, expressed as estimated changes per week treatment to account for variation between patients in the timing of the PET scan/CT response evaluation. Patients (n = 19) are represented by two bars (blue and pink) and grouped per best overall treatment response. Blue bars, change in sum target lesions according to RECIST between pretreatment and first response evaluation. Pink bars, average SUVmax change. Dots are individual lesions (n = 111). Individual lesion datapoints for size (blue) and uptake (red) are connected by gray lines. Blue dots, lesion blueness, RECIST diameter pretreatment. Dot location, change in size versus baseline. Red dots, lesion redness, SUVmax pretreatment. Dot location, SUVmax change.

Uptake in tumor lesions during treatment

During treatment, the average 89ZED88082A uptake in nonirradiated lesions in all patients (lesion n = 111) was lower compared to baseline (−4.6% change in geometric mean SUVmax per week of treatment, 95% CI −6.5% to −2.6%), a change that depended on best overall response with a greater decrease in patients with SD, PR or CR (Pinteraction = 0.018) (Extended Data Fig. 6c). Of the eight patients who showed PR or CR on treatment, five already met criteria for PR at the time of the PET scan at 30 days. When taking into account tumor volume change and resulting tracer uptake underestimation due to partial volume effects in responding lesions, the estimated average tracer uptake change was −2.7% (95% CI −4.4% to −1.1%) per week treatment, which no longer depended on best overall response (Pinteraction = 0.71) (Extended Data Fig. 6d). No patient in the repeat imaging cohort experienced pseudoprogression.

Within patients, lesions demonstrated diverse changes in 89ZED88082A uptake, with some decreasing and others increasing compared to baseline. Moreover, responding lesions displayed a variety of dynamics in 89ZED88082A-uptake change between the two PET series (Fig. 4c).

For ten patients, paired tumor tissues of the same lesion with corresponding tumor volumes of interest (VOIs) on PET were available (Supplementary Information Fig. S1). Five of them reflected concordant treatment-emergent changes by IHC and imaging (Fig. 3b). In one patient, a lymph node metastasis with a SUVmax of 8.28 and stromal CD8 T cell infiltration at baseline showed only normal lymph node tissue in the second biopsy sample, with SUVmax of 5.63 on the on-treatment PET.

Normal tissue biodistribution and pharmacokinetics

89ZED88082A showed a specific uptake per organ (Fig. 1). The highest 89ZED88082A uptake occurred in the spleen and was apparent within an hour of injection. From day 2 onwards, there was a clear 89ZED88082A uptake in normal lymphoid tissues, including the bone marrow, Waldeyer’s ring, lymph nodes, the small intestine (Extended Data Fig. 1) and the appendix (Extended Data Fig. 7i). Sites with previous lymph node dissection lacked uptake. Furthermore, tracer uptake was present in the renal cortex and liver. Partial volume effects and spillover signal precluded the quantification of small tumor lesions contained within the renal cortex and the spleen. Tracer uptake was also observed at sites of inflammation (Extended Data Fig. 7). In two patients, 89ZED88082A uptake was lower in vertebrae irradiated <12 months earlier than in nonirradiated vertebrae (Extended Data Fig. 7g,h). During treatment, the average tracer SUVmean in blood pool at four weeks was 13.3% lower compared to pretreatment. Equally, uptake in spleen and lymphoid tissues was limitedly decreased, the latter not being correlated to best overall response (Extended Data Fig. 1b).

Several patients developed immune-related AEs (irAE) after ICI initiation (Extended Data Table 1). One patient with Hashimoto’s thyroiditis on stable thyroid replacement therapy experienced a flare-up requiring more replacement. Her elevated baseline thyroid SUVmean of 3.32 increased during treatment to 8.07 (Extended Data Fig. 7e,f). In other patients experiencing irAE ≥ grade 3 within the time frame of PET scans or thereafter, no higher 89ZED88082A uptake at baseline or during treatment occurred in organs of interest. This included two patients who developed diarrhea 4 and 14 days after the on-treatment CD8 PET. They were evaluated two days after the start of diarrhoea with colonoscopy and a colonic biopsy, which showed minor inflammation in both patients. They were later treated with steroids because of clinical suspicion of ICI-induced colitis.

In part A, serum 89ZED88082A/CED88004S protein levels were comparable within the same dose groups (Extended Data Fig. 8a,b). The estimated serum half-life of 89ZED88082A/CED88004S was 1.19 ± 0.33 days. Tracer pharmacokinetics were not influenced by ICI (Extended Data Fig. 8c). 89ZED88082A was intact in serum, while only low molecular weight components, including free 89Zr, were detectable in urine (Extended Data Fig. 8d). 89ZED88082A administration did not affect T cell, B cell and NK cell blood counts (Extended Data Fig. 3a).

No patient had endogenous antibody-drug antibodies (ADAs) before tracer injection (n = 31), 19% developed ADAs 28–50 days after the first (n = 5 out of 26) and 8% 18–38 days after the second tracer injection (n = 1 out of 12). One out of the 22 patients imaged twice (pre- and on treatment) developed ADAs after the first tracer injection. There was no apparent ADA effect on 89ZED88082A/CED88004S serum levels and imaging results.

Discussion

A systemic characterization of the tumor microenvironment is critical for understanding an effective anticancer immune response following immunotherapies. This is a first-in-human study with the CD8-targeting antibody 89ZED88082A characterizing the CD8+ T cell biodistribution by PET imaging in patients with cancer at baseline and during ICI treatment. We demonstrated that the tracer is safe. Tracer uptake in tumor lesions correlated with CD8 IHC and autoradiography signal in those lesions. 89ZED88082A signal was conspicuous early on in the blood pool and kidneys as clearance organs, and in the spleen with extensive CD8 expression on the red pulp reticuloendothelial cells18. However, progressive uptake was evident only in CD8-rich tissues such as the lymph nodes, further supporting the tracer’s CD8 specificity.

Overall, high 89ZED88082A tumor uptake at baseline was associated with a better OS, concordant with findings from CD8 IHC in tissues from clinical ICI trials6,19. There was a major spatial heterogeneity within and between patients in 89ZED88082A uptake by their lesions. We took two approaches to verify whether potential differences in CD8 tracer uptake did reflect CD8-related tumor characteristics. First, we showed higher tracer uptake in dMMR than in pMMR tumors imaged before treatment, reflecting the higher CD8+ T cell infiltrate reported in dMMR tumors20,21,22,23,24. Second, we showed that tumor lesions biopsied and known by IHC to have a high T cell infiltrate (either ‘stromal’ or ‘inflamed’ phenotype) showed higher CD8 tracer uptake than the group with a low-T cell ‘desert’ phenotype. The 89ZED88082A uptake in a rim pattern in several tumors before and during treatment likely mirrors CD8+ T cell tumor infiltration referred to as the invasive margin11,23,25.

To improve insight into ICIs, their biodistribution has been studied with 89Zr-labeled anti-PD-1 and anti-PD-L1 antibodies15,16,26,27. In patients receiving atezolizumab, pretreatment 89Zr-atezolizumab tumor uptake predicted tumor response, PFS and OS, while PD-L1 expression assessed by IHC did not15. Similar observations were made for 89Zr-pembrolizumab imaging16. This demonstrates that T cells in tumor lesions as key mediators of immunotherapy can be evaluated by whole-body PET imaging. CD8 imaging was recently described in a small phase 1 study involving CD8 PET imaging at a single time point either before, during or after ICI or targeted therapy in 15 patients using different protein doses of the minibody 89Zr-Df-IAB22M2C28. The 89Zr-minibody was safe and accumulated in CD8+ rich tissues and tumor lesions of ten patients, supporting the CD8 PET approach.

Although we observed increasing signal in individual cases preceding a response, as also shown in some biopsy studies9,10,11,29, overall SUVmax changes on 89ZED88082A PET at 30 days after initiation of ICI did not correlate with best overall response when adjusted for volume changes. Intriguingly, we identified an enormous interlesional heterogeneity in tracer uptake on PET at 30 days in patients who responded. These findings indicate a remarkable spatio-temporal variability in systemic T cell dynamics as an antitumor immune response unfolds. Interestingly, similar results have been seen in a well-controlled mouse model using in situ fluorescent imaging of tumor cells and immune cells. Thus, a large variety in immunophenotype evolution was visualized even within individual mice of one model of the same seeded tumor cell line30. Moreover, in a human tumor fragment platform assay, PD-1 blockade resulted in different immune activation profiles among small tumor fragments derived from individual patient tumor lesions31. Together, our results underscore the importance of timing and characterization of all tumor lesions in comprehensively evaluating the tumor-immune status and therapy-induced pharmacodynamic effects.

Some tumor types display faster response kinetics to ICIs than others32,33. At 30 days, we captured a snapshot of patients and their lesions at different stages of their immune response, or lack thereof. Our results indicate that earlier imaging time points are warranted to capture CD8+ T cell dynamics that may be preceding the antitumor activity resulting in lesion shrinkage in these patients. Since various tumor types were included in our study, the numbers of individual tumor types enrolled were too small to define patient subset-specific CD8+ T cell kinetics. To fully understand and assess antitumor immunity induced by ICIs beyond what is feasible with localized tumor biopsies, it is essential to image T cell dynamics across lesions by whole-body evaluation over time. Because 89Zr has a relatively long half-life of 78.4 h, repeated PET imaging with 89Zr tracers ideally requires an interval of two weeks to avoid residual radioactivity and allow full clearance of the antibody. New small molecule tracers targeting CD8 and labeled with fluorine-18 may more readily allow sequential imaging time points, increasing the chance of capturing a more complete time course, to elucidate spatio-temporal changes in CD8+ T cells following initiation of immunotherapy34. For future studies, we envision also an earlier second imaging time point, namely within two weeks after starting ICI therapy, to capture pharmacodynamic changes before substantial tumor shrinkage.

Several issues challenged the interpretation of CD8 imaging changes following treatment. The uptake pattern changed rather than the magnitude of uptake in some tumor lesions, probably reflecting enhanced infiltration in a larger tumor volume. We expressed specific tumor uptake as SUVmax, commonly used to measure specific uptake. However, this may not properly reflect heterogeneous uptake or a change in distribution pattern.

In addition, we detected CD8+ T cells in areas of nonmalignant inflammation, supporting the tracer’s ability to visualize inflammatory processes in any setting including 89ZED88082A PET changes during ICI treatment in a patient with Hashimoto’s thyroiditis, a disease with high lymphocyte involvement35. Therefore, CD8 PET may identify potential irAEs if patients are scanned in the relevant time frame. However, it should be noted that not all irAEs are driven by CD8+ T cells, and instead may involve multifactorial aetiologies including B cell, complement or auto-antibody driven mechanisms36. Thus, the potential relevance of CD8 PET in the characterization, identification and monitoring of irAEs will require further study and is currently limited to a single anecdote.

The tracer showed an organ-specific biodistribution in normal tissues without in vitro signs of cellular tracer internalization by immune cells. We cannot exclude that we also visualized CD8+ NK cells, but they are relatively rare and not likely to be confounding. Uptake in the spleen was conspicuous within the first hour postinjection, likely due to high perfusion and facile access of the tracer to high CD8 levels by littoral cells lining the red pulp sinusoids15,18. The higher spleen 89ZED88082A SUVmean at 4 mg than at 10 mg likely reflects partial CD8 saturation at the 10 mg dose, due to containing more unlabeled CED88004S.

High bone marrow uptake early after injection, followed by a gradual decline in this densely vascularized space, is likely related to perfusion, while imaging at later time points likely reflects target-mediated 89ZED88082A binding to CD8+ T cells, which would be expected based on its role as a primary and secondary lymphoid organ and memory CD8+ T cell localization37,38. Moreover, we saw tracer uptake in the small intestine, likely showing CD8+ T cells in the gut-associated lymphoid tissue, such as the Peyer’s patches within the gut mucosa39,40. High tracer uptake in these tissues matched sites of CD8 protein expression reported in the Protein Atlas41, although these comparisons cannot be exact due to the relatively young and healthy sources of tissues in the atlas, and the relative complexity of delivering antibody tracer to the CD8 target in living subjects. Tracer signals in liver, renal cortex, urine and large bowel probably reflected tracer clearance and metabolism rather than target-mediated binding. The renal cortex showed a persistent high radioactive signal irrespective of decreasing blood pool levels. This is presumably due to renal tracer clearance followed by resorption and catabolism with residualization of intracellular charged metal chelate catabolites such as lysine-DFO-Zr-binding proteins. This is a known phenomenon for small molecules and antibody fragments42,43.

Serial, whole-body characterization of CD8+ T cells has several potential applications in clinical research. One application is to more fully characterize the pretreatment CD8+ T cell tumor infiltration, which may function as a predictive biomarker for subsequent response to a particular immunotherapy (for example, ICIs). Furthermore, serial CD8 PET imaging has the potential to characterize treatment-emergent pharmacodynamic changes following new immunotherapies or combinations of agents, and may therefore prove useful in guiding their clinical development. 89ZED88082A PET may also be helpful to guide tumor biopsies to improve the chance of obtaining a tumor sample with high CD8+ T cell infiltration. Ultimately, CD8 PET has the potential to become a clinical decision support tool to individualize immunotherapeutic approaches in patients. Describing and accepting the huge spatial and temporal heterogeneity of CD8+ T cells is critical towards a more individualized treatment approach in the future. However, the generation of much larger CD8 PET imaging data sets and correlation with clinical outcomes will be needed to assess whether CD8 PET can guide treatment decisions.

In conclusion, 89ZED88082A PET specifically visualizes CD8 in vivo, offering the opportunity to assess whole-body CD8+ T cell distribution, not obtainable with a single-lesion biopsy. We demonstrated that CD8+ T cell presence in tumor lesions imaged before ICI could be predictive for OS, highlighting the potential of CD8 imaging as a predictive biomarker to personalize treatment for patients. The dynamics of intratumoral CD8 expression during ICI exposure is more complex and nuanced than previously reported and differs between patients and lesions in the same patient. To properly evaluate tumor-immune status, timing and evaluation across lesions are crucial. Our results provide a strong rationale to characterize the tumor-immune microenvironment using new imaging technologies.