Kinetic analysis of HER2-binding ABY-025 Affibody molecule using dynamic PET in patients with metastatic breast cancer

The introduction of trastuzumab, a HER2-targeting monoclonal antibody, improved the outcome of HER2-enriched breast cancer treatment dramatically [1, 2]. Still, breast cancer is a heterogeneous disease and HER2 expression may vary considerably not only within a given lesion but also between the primary tumour and metastases [3‐5]. Obtaining multiple biopsies in a patient is not always possible, and the need has emerged for a robust imaging technique to estimate HER2 status in cancer patients that is non-invasive, quantitative, and cost-effective and allows for repeated examinations.

One of the most successful classes investigated as HER2-imaging agents is Affibody molecules [6]. The second-generation anti-HER2 Affibody molecules (code-named ABY-025) have high stability, diminished interaction with immunoglobulins, and picomolar affinity for HER2 as well as rapid pharmacokinetics with fast blood clearance, resulting in high tumour-to-background ratios shortly after injection [7‐10]. In a pilot study, ⁶⁸Ga-ABY-025 PET uptake quantification suggested a preliminary SUV threshold of 6 for the discrimination of HER2-positive from HER2-negative lesions [11, 12]. SUV is a highly repeatable measurement; however, it relies heavily on the acquisition and reconstruction protocols used [13].

SUV correlated strongly with IHC scores performed on biopsies. However, the liver had the second highest SUV values among normal tissue after the kidneys at 60-240 min post-injection with about 9% of the injected activity retained by the liver [12, 14]. The relatively high liver background uptake might mask the signal from smaller liver metastases, especially those with low to moderate HER2-receptor expression, and that might compromise the accuracy of HER2 scoring by PET. The solution for the improved image analysis in the liver is crucial given that HER2-positive breast cancer subtypes have a higher probability of liver metastasis which, in these subtypes, is the second most common metastatic site, compared to HER2-negative subtypes [15].

SUV does not distinguish between specific and non-specific uptake in receptor-binding radiotracers such as, in this case, ⁶⁸Ga-ABY-025 [16, 17]. A more accurate investigation of the underlying uptake properties requires kinetic modelling. The rate constants obtained from fitting the time-activity curves (TACs) using kinetic modelling can potentially be used to increase contrast by eliminating non-specific background activity found in surrounding tissue. This, in principle, could enhance tumour visualization in these organs [18, 19]. Individual rate constants, or micro-parameters, describing the rate of transport of the tracer between plasma and tissue and between different states (e.g. free and internalized) in tissue can be estimated using compartment models. The gold standard for tracer kinetic analysis is the estimation of these rate constants by non-linear regression of the operational equations describing these compartment models. In case of a tracer with irreversible kinetics, these micro-parameters can then be used to estimate the net influx rate of the tracer, which, in the present case, is assumed to be related to the HER2. Compartment models can be simplified by linearization using basis functions, which allows for a fast calculation of rate constants for each voxel, resulting in parametric images showing the parameter of interest at the voxel level. A simplified linearized method is the Patlak method, which is a graphical analysis technique that allows for the estimation of the net influx rate without giving information on the individual rate constants.

The present study aimed to (1) explore the applicability of kinetic modelling and (2) parametric image analysis for absolute quantification of ⁶⁸Ga-ABY-025 uptake and HER2-receptor expression and (3) how that relates to static SUVs.

The kinetics of ⁶⁸Ga-ABY-025 were best described by the irreversible two-tissue compartment model (2TC-3k model), giving the lowest Akaike information criterion in 22 out of 40 TACs (55%). Patlak K_i values showed an excellent agreement with 2TC K_i values both for VOI-based analysis and in parametric images (Fig. 3a, b). Both Patlak and 2TC K_i values derived from parametric images agreed very well to their counterparts obtained from the VOI-based analysis (n = 24, R² > 0.99, and p < 0.001 for both correlations; Fig. 3c, d).

K_i presented good contrast and low background uptake in the normal liver (Fig. 4a). K_i values were 3.7- and 7.1-fold higher in the metastatic lesions compared to the normal liver (Tmax/Nmean ratio 2TC, 3.7 ± 2.8; Patlak, 7.1 ± 7.8). The Patlak Tmax/Nmean ratio was significantly higher (p < 0.05) than the corresponding SUV-based ratio (4.2 ± 3.4 at 2 h post-injection). All metastases invariably had lower tracer delivery rates than normal liver and were visualized as cold spots in K₁ images (Fig. 4a, b). Parametric image-based rate constant values and SUV for both liver metastases and normal liver are presented in Table 2.

⁶⁸Ga-ABY-025 binding in HER2-negative lesions was equal to or lower than the uptake in the normal liver in both K_i and static SUV images (Fig. 4b). K_i values of 0.015 mL/cm³/min for 2TC-3k model and 0.013 mL/cm³/min for the Patlak model corresponded to the proposed SUV cut-off of 6.0 at 2 h post-injection [12], yielding 95% confidence intervals of 4.4-7.6 (2TC) and 4.8-7.2 (Patlak).

Static SUVs obtained at 2 h and 4 h post-injection had a good correlation with parametric image-based K_i (Patlak R² = 0.95; 2TC R² = 0.87, n = 9, Fig. 5). T/N ratios were 4.2 ± 3.4 and 4.7 ± 4.6 for 2 h and 4 h post-injection, respectively (n = 10). ⁶⁸Ga-ABY-025 PET test-retest reliability was high as shown by the parametric image-based K_i values in the test-retest group of patients (Pearson r ≥ 0.92, n = 5) with a relative repeatability coefficient of 30% for Patlak K_i and 47% for 2TC K_i compared to 32% for SUV at 2 h (Fig. 6). Neither systematic nor proportional bias was observed in the test-retest group (Fig. 7).

In this study, we assessed the kinetics of ⁶⁸Ga-ABY-025 with the hypotheses that visualization and determination of HER2 status in liver metastases could be facilitated by eliminating background uptake through creating parametric images of the tracer uptake kinetics, and that simple static SUVs in tumours are acceptable surrogates for the true binding kinetics of the tracer.

The 2TC-3k irreversible model best fitted the data according to the Akaike information criterion. This coincided with the preclinical findings showing almost irreversible kinetic properties of Affibody molecules in HER2-expressing tumours [22]. Since the majority of the measurements in patients were taken from liver lesions, the heterogeneity induced by spill-in from gradually lower uptake in normal liver probably explains why there was a small reversible component in a subset of the measurements.

Both VOI-based kinetic modelling and parametric image analysis returned virtually identical K_i values (R² = 0.99, n = 24), indicating that VOI-based kinetics are reproducible on the voxel level. The basis function implementation of the 2TC-3k model also provided K₁ images that reflected the delivery rate of the tracer to various tissues, including tumours. This is important when estimating the fraction of the tracer available to cancer lesions for the subsequent computation of the tracer net influx rate K_i.

There was a strong correlation between tumour static SUV values and K_i values (Patlak R² = 0.95; 2TC R² = 0.87, n = 9, Fig. 5), indicating that the static SUV images provided a close representation to HER2-specific binding since parametric K_i images reflect the specific uptake solely and eliminate any non-specific binding that increases background uptake. The relatively narrow 95% confidence intervals, especially with Patlak K_i, around the proposed SUV cut-off of 6 uphold its use in the upcoming clinical trials and provide a reference to the degree of uncertainty paired with it. Moreover, there was a very good correlation in the test-retest group (Fig. 6), with a relative repeatability coefficient of 30% for Patlak model, indicating high reproducibility of kinetic analysis and similar to the results with static SUV values demonstrated previously [12]. These findings further support the claim that SUV values can be used for the quantification of HER2 expression, as concluded in our previous paper [12]. This will be used in a large multicentric phase II/III clinical trial for non-invasive quantification of HER2 expression in advanced breast cancer using ⁶⁸Ga-ABY-025 PET (NCT03655353) that is currently ongoing.

A noticeable degree of specific uptake was recognized in normal liver, further supporting the claim that HER2 receptors are expressed in normal liver tissue [23], although at levels significantly lower than in HER2-positive tumours. The Human Protein Atlas describes liver HER2 expression as moderate, compared to other normal tissues [24]. Liver retention in static PET images is much higher than what can be accounted for by specific binding. This was explained by the very high first-pass extraction with K₁ values significantly higher than for all tumours, probably related to the specialized non-continuous endothelium of liver capillaries. While visualization of the HER2-expressing lesions was possible in both parametric K_i and static SUV images (Fig. 4a), with the feasibility to localize small liver metastases, tumour-to-liver ratios were significantly higher in parametric Patlak K_i images. Thus, parametric imaging might become beneficial in a few clinical cases where it might be difficult to grade the HER2 expression in small metastases due to the high liver background uptake. Because dynamic imaging introduces an additional 45-min scan, the presence of small suspicious liver lesions (< 1 cm) should be known from other imaging modalities beforehand and the resulting HER2 score should be of direct relevance for therapeutic decisions. In terms of contrast, the Patlak method performed better than the 2TC-3k method, mainly because K_i values in the normal liver were 30% lower with the Patlak method than with the 2TC-3k model, while the Patlak distribution volume (V_e) values were 30% higher than those based on the 2TC-3k model (V_ND). The kinetics of ⁶⁸Ga-ABY-025 in the liver are not irreversible, which leads to an underestimation of K_i and an overestimation of V_e when using the Patlak method, and an additional reason of the higher Patlak distribution volume is that it includes blood volume, whereas V_ND does not. Hence, the improved contrast using Patlak K_i is essentially a fortunate benefit of the erroneous assumption of irreversible kinetics in healthy liver tissue, likely resulting in better lesion detection in Patlak K_i images than in SUV images, as clearly shown in Fig. 8. Most peptide-based PET tracers are retained and metabolized in the liver, and as a result, non-specific liver uptake is a common problem in PET imaging. This phenomenon is also observed in recent biodistribution studies [25]. Similarly, preclinical studies investigating the use of other Affibody molecules targeting HER3 receptors also observed non-specific uptake of the given tracer in the liver [26]. Parametric image formation with the enhanced tumour-to-liver ratio is a favourable workaround to this issue not limited to ABY-025 PET but could potentially establish a framework for future studies with similar tracers.

HER2 imaging provides an extended scale to the degree of expression contrary to traditional IHC that incorporates the entire range of positive lesions into the ordinal HER2 3+, as shown in the previous studies [11, 12, 27].

The main limitation of the study is the small number of subjects. Data was used to explore multiple aims, which increases the risk of type 1 errors. There were few lesions with a definite diagnosis from biopsies, and only few patients had HER2-negative disease status. A substantial fraction of lesions were small, which introduces measurement errors. Dynamic studies in the present work were limited to a single bed position, allowing for only 15 cm axial field of view. Multi-bed dynamic PET in current clinical scanners or research total-body PET scanners with automated definition of the input function could allow for performing whole-body parametric imaging and correlative analysis of SUV and T/B ratios towards parametric imaging of the entire body [28, 29]. Future studies should also investigate if the dynamic protocol could be shortened.

Parametric imaging provided good visualization and eliminated non-specific background uptake in the liver. There was a good correlation of absolute quantification presented by kinetic parameters and the simple SUV from static 2 h and 4 h time point images, further supporting the use of SUV for clinical routine in whole-body imaging and HER2-expression quantification. Besides, K₁ and K_i images provided insight on the pharmacokinetics of ⁶⁸Ga-ABY-025. K_i values might be useful for the quantification of HER2 expression in breast cancer patients under certain conditions, such as anti-HER2 drug development studies and in some clinical cases with known small tumours in the liver. Further studies with a larger patient cohort and more biopsies are warranted.