The goal of the study was to investigate if parameters derived from 18F-FDG PET imaging and IHC, singularly or in combination, could reliably distinguish different human HNC xenograft models from one another. Eventually, this could give direction to classification methods for clustering of tumors that are most alike regarding multiple characteristics in clinical studies, e.g. for treatment prediction and prognostication purposes or for individualized treatment selection.
IHC markers were selected for relevance in metabolic cell processes and known therapy resistance mechanisms , as well as for (in)direct links to 18F-FDG tumor uptake in the literature [34–38]. Using a systematic analysis method, the presented results show that a finite set of IHC staining parameters, quantifying several relevant molecular cell processes, can accurately allocate a specific tumor to the appropriate tumor line within a cluster of 14 HNC lines. Adding more staining markers increases accuracy, but at a certain point this effect levels off. A specifying accuracy of at least 70% can be achieved with a random set of 6 of these IHC markers.
18F-FDG PET could not differentiate between the HNC lines in this study. Furthermore, quantification parameters (SUV, T/M) and selected 18F-FDG PET texture features did not provide additional value to classification accuracy by IHC alone. It may be unlikely that 18F-FDG PET derived parameters can reliably categorize combined differences in biological characteristics between head and neck tumors. Absolute SUVs were relatively low in this study and were in line with other preclinical HNC studies [39, 40], but lower than the typical SUVs that are detected in clinical HNC . This is inherent to the mouse model used for PET imaging in this study. Although differences were seen between HNC lines, most of the observed variance could be attributed to intra-tumor line differences.
Uptake of 18F-FDG has been assessed for correlation with several biological markers in tumors, such as GLUT1, glycolysis- and hypoxia-related markers [34, 35, 42], proliferation [36, 42, 43], EGFR  and AKT , with conflicting results. Overall, 18F-FDG uptake in malignancies reflects multifactorial mechanisms of increased metabolic activity and glucose utilization, performed by glucose transporters and enzymes in the glycolytic pathway, which in turn are regulated through different signaling pathways triggered by endogenous and exogenous stimulators. Aims to attribute 18F-FDG uptake to expression of one specific protein or therapy resistance mechanism are therefore debatable.
Quantitative texture feature analysis has been introduced in radiodiagnostic imaging as a means to characterize and classify tumors using their signal intensity distribution [44, 45], and studies described the use of texture features as potential prognostic or predictive tools [46, 47]. Textural feature analysis can be applied in numerous imaging modalities where lesion configuration plays a discriminating role for stratification , e.g. contact dermoscopy images  or microscopy images [50, 51]. For this study we focused on a limited set of global features that would give relevant insight in signal distribution next to quantification parameters such as IHC staining fraction or PET SUV, including entropy and skewness for IHC and PET images, with the additional feature “mean” (pixel grey value) for IHC images. IHC texture features combined with IHC parameters conveyed optimal characterization accuracy. However, addition of 21 feature values improved the classification accuracy of the combined 9 IHC parameters (which was already 74.9%) by only 4.9%.
Limitation of the study is the use of relatively small xenograft tumors as opposed to multiple biopsies from larger HN tumors. However, this setup provides the possibility to study multiple parameters in entire tumor sections, which is difficult to achieve on a large scale in a patient setting. In clinical studies, sampling errors by extraction of a single biopsy forms a general pitfall when assessing biological markers with a heterogeneous tumor distribution. At least 4-5 central core biopsies are needed to minimize effects of IHC staining heterogeneity within tumor sections [11, 52]. In entire tumors, an even greater spatial heterogeneity in IHC characteristics is likely to occur. Iakovlev et al. demonstrated that, for CA9 quantification in multiple cervical tumor biopsies per patient, the highest variation was inter-tumor, followed by intra-tumor and intra-tumor section variation. The greatest reduction in assessment-error could be achieved by increasing the number of biopsies spaced well apart rather than increasing the number of stained sections per biopsy .
We analyzed multiple tumors per xenograft model, which have the same genetic background and are grown to a similar size under similar circumstances in mice from the same strain. Even these tumors, that may represent a basic approach to multiple biopsies from heterogeneous tumors in different patients, exhibited variable characteristics during growth, affected by microenvironmental and external factors [53, 54]. Intra-tumor line variation for the administered exogenous markers was overall larger than for endogenous markers. Tumor uptake of exogenous markers is influenced by dosage and administration, circulation and body clearance properties, tumor vascular density and perfusion, diffusion, binding and washout kinetics et cetera. In the clinical situation, external and microenvironmental influences may result in even larger intra-tumor and inter-tumor variation of molecular marker expression in HNC.
Results from the study can be extrapolated to other tumor types in the sense that, when the aim is to allocate or adapt individually tailored treatment, a selection of parameters provides the potential for precise tumor characterization and stratification. Depending on the treatment options at hand, individual tumor profiles or grouping of most uniform tumors can be established with the help of a distinct panel of IHC markers. This precludes analyzing an extensive number of classification parameters.
Care should be taken that the number of chosen characterizing parameters is not too small either. In this study, we found relatively low accuracies when less than 6 IHC parameters were combined for classification. Instead of administering exogenous IHC markers, molecular PET tracers with a more defined imaging spectrum than 18F-FDG, such as tracers for hypoxia or proliferation rate , can potentially complement IHC analyses by visualizing the entire tumor for presence of certain tumor mechanisms relevant for treatment.