In Situ Metabolomics Expands the Spectrum of Renal Tumours Positive on 99mTc-sestamibi Single Photon Emission Computed Tomography/Computed Tomography Examination

Background Definite noninvasive characterisation of renal tumours positive on 99mTc-sestamibi single photon emission computed tomography/computed tomography (SPECT/CT) examination including renal oncocytomas (ROs), hybrid oncocytic chromophobe tumours (HOCTs), and chromophobe renal cell carcinoma (chRCC) is currently not feasible. Objective To investigate whether combined 99mTc-sestamibi SPECT/CT and in situ metabolomic profiling can accurately characterise renal tumours exhibiting 99mTc-sestamibi uptake. Design, setting, and participants A tissue microarray analysis of 33 tumour samples from 28 patients was used to investigate whether their in situ metabolomic status correlates with their features on 99mTc-sestamibi SPECT/CT examination. In order to validate emerging data, an independent cohort comprising 117 tumours was subjected to matrix-assisted laser desorption/ionisation mass spectrometry imaging (MALDI MSI). Outcome measurements and statistical analysis MALDI MSI data analysis and image generation were facilitated by FlexImaging v. 4.2, while k-means analysis by SCiLS Lab software followed by R-package CARRoT analysis was used for assessing the highest predictive power in the differential of RO versus chRCC. Heatmap-based clustering, sparse partial least-squares discriminant analysis, and volcano plots were created with MetaboAnalyst 3.0. Results and limitations We identified a discriminatory metabolomic signature for 99mTc-sestamibi SPECT/CT–positive Birt-Hogg-Dubè–associated HOCTs versus other renal oncocytic tumours. Metabolomic differences were also evident between 99mTc-sestamibi–positive and 99mTc-sestamibi–negative chRCCs, prompting additional expert review; two of three 99mTc-sestamibi–positive chRCCs were reclassified as low-grade oncocytic tumours (LOTs). Differences were identified between distal-derived tumours from those of proximal tubule origin, including differences between ROs and chRCCs. Conclusions The current study expands the spectrum of 99mTc-sestamibi SPECT/CT–positive renal tumours, encompassing ROs, HOCTs, LOTs, and chRCCs, and supports the feasibility of in situ metabolomic profiling in the diagnostics and classification of renal tumours. Patient summary For preoperative evaluation of solid renal tumours, 99mTc-sestamibi single photon emission computed tomography/computed tomography (SPECT/CT) is a novel examination method. To increase diagnostic accuracy, we propose that 99mTc-sestamibi–positive renal tumours should be biopsied and followed by a combined histometabolomic analysis.


Introduction
Recent advances in genomics and molecular genetics have provided novel insights into renal tumorigenesis refining the molecular classification of renal cancer [1]. In addition to providing a better understanding of the molecular landscape of major renal cell carcinoma (RCC) subtypes, a pan-genomic study from The Cancer Genome Atlas Research Network has led to a more accurate definition of their biological behaviour [2]. A subset of metabolically divergent RCCs was identified, displaying a distinct metabolic expression associated with extremely poor survival [2]. This has confirmed the recently documented correlation of metabolic expression subtypes and patient survival across various and diverse malignancies [3].
Imaging with 99m Tc-sestamibi single photon emission computed tomography/computed tomography (SPECT/CT) was recently introduced for preoperative RCC diagnosis by dividing solid renal tumours into positive or negative ones on the basis of tracer uptake [4]. As a mitochondrial agent, sestamibi uptake correlates with different mitochondrial content and variable multidrug resistance pump expression that renal tumours exhibit [5]. The 99m Tc-sestamibipositive tumours exhibiting increased 99m Tc-sestamibi uptake are more likely to be benign or of low malignant potential, whereas the 99m Tc-Sestamibi-negative counterparts appear to have malignant characteristics. The latter group could be considered for surgery, while the former could potentially be managed conservatively by active surveillance utilising long follow-up with or without renal biopsy [6].
Technological advances of molecular imaging and pathology are expected to reshape modern medicine. For example, a diagnostic differentiation of renal oncocytomas, a benign tumour, from malignant renal tumours appears promising on imaging grounds. Nevertheless, false-positive and false-negative results on 99m Tc-sestamibi SPECT/CT confound certain aspects of its clinical utility. Herein, we investigate the in situ metabolomic status of a 99m Tcsestamibi SPECT/CT-examined cohort of renal tumours and propose an integrated approach that combines molecular imaging and in situ metabolomic profiling to better characterise renal neoplasia and potentially improve patient management.

2.
Patients and methods

Case selection
Forty-two out of the 50 patients who were included in the current study participated previously in an institutional review board-approved prospective study to investigate imaging characteristics of solid renal tumours (T1; 7 cm) using 99m Tc-sestamibi SPECT/CT [6]. The results of the investigation were reported in two consecutive studies [7,8] clustering, sparse partial least-squares discriminant analysis, and volcano plots were created with MetaboAnalyst 3.0. Results and limitations: We identified a discriminatory metabolomic signature for 99m Tc-sestamibi SPECT/CT-positive Birt-Hogg-Dubè-associated HOCTs versus other renal oncocytic tumours. Metabolomic differences were also evident between 99m Tc-sestamibi-positive and 99m Tc-sestamibi-negative chRCCs, prompting additional expert review; two of three 99m Tc-sestamibi-positive chRCCs were reclassified as low-grade oncocytic tumours (LOTs). Differences were identified between distal-derived tumours from those of proximal tubule origin, including differences between ROs and chRCCs.
Conclusions: The current study expands the spectrum of 99m Tc-sestamibi SPECT/ CT-positive renal tumours, encompassing ROs, HOCTs, LOTs, and chRCCs, and supports the feasibility of in situ metabolomic profiling in the diagnostics and classification of renal tumours. Patient summary: For preoperative evaluation of solid renal tumours, 99m Tcsestamibi single photon emission computed tomography/computed tomography (SPECT/CT) is a novel examination method. To increase diagnostic accuracy, we propose that 99m Tc-sestamibi-positive renal tumours should be biopsied and followed by a combined histometabolomic analysis.

TMA construction and MALDI MSI analysis
To investigate the in situ metabolomic status, tumour samples from biopsy and resection specimens were arranged in a tissue microarray analysis (TMA) format using a semiautomated tissue arrayer MiniCore.
For each tumoural case, representative areas were selected and marked on an HE-stained slide. Accordingly, three cores (for resections) and/or one core (for biopsies) with a diameter of 1 mm were extracted from the "donor" block and brought into the "recipient" paraffin block. To validate emerging data, we used part of a previously published cohort [9],

3.1.
Clinicopathological characteristics of 99m Tc-sestamibi SPECT/CT-analysed renal tumours and initial metabolomic data acquisition The examined cohort comprised a total of 33 renal tumours from 28 patients ( Table 1). Seven of nine ROs were positive, while two of nine were negative on the 99m Tc-sestamibi SPECT/CT examination. All three HOCTs identified in a female patient with verified Birt-Hogg-Dubè (BHD) syndrome were also positive. The BHD patient had a germline FLCN heterozygous mutation (c.779 + 1G > T), repeated episodes of pneumothorax, and cutaneous basal cell carcinomas. Three of seven chRCCs exhibited a positive 99m Tc-sestamibi uptake, whereas the remaining four chRCCs were negative ( Table 2). All evaluated ccRCCs (6/ 6) and pRCCs (8/8) were negative on the 99m Tc-sestamibi SPECT/CT examination.
Overall, approximately 770 individuals' MS peaks per pixel within the mass range of m/z 100-1000 could be resolved within the tissue examined, while 319 metabolites were annotated through the HMDB.

3.2.
Hierarchical clustering analysis of metabolomic data segregates positive BHD-associated HOCTs and distinguishes LOTs from classic chRCCs Unsupervised hierarchical clustering analysis identified a discriminatory metabolomic signature for positive BHD- associated HOCTs. This is in accordance with the recent molecular evidence suggesting that an HOCT represents an entity with genomic features intermediate between an RO and a chRCC [14]. All HOCTs as well as the three ROs arising in one patient were subclustered together (Fig. 1A). Metabolomic differences were also found between positive and negative chRCCs (Fig. 1B), prompting an expert review of the morphological and immunohistochemical features of the set of cases initially considered "chRCCs". Two of three positive chRCCs were reclassified upon review as "lowgrade oncocytic tumours" (LOTs). A LOT is an emerging renal entity with morphological features overlapping those of an RO and a chRCC that demonstrates a CK7 pos./CKIT neg. immunoprofile, and lacks the multiple chromosomal losses typically seen in chRCCs [15]. The third positive chRCC was classified as an eosinophilic chRCC upon review (Fig. 1C). This approach reduced the total number of positive chRCCs to one, and hence further comparison of positive ROs versus chRCCs was precluded. However, the tumour spectrum of renal neoplasms exhibiting uptake on 99m Tcsestamibi SPECT/CT examination was expanded (Fig. 1D), with potential implications for clinical management.

Metabolic alterations in RCC subtypes with regard to the presumed origin
Similar to other studies on metabolomic and gene expression profiling [16], we confirmed the differences between the distal nephron-derived tumours (ie, chRCCs) from those originating from the proximal tubules (ie, ccRCCs and pRCCs), and highlighted others within the same subgroups. Annotated metabolites emerging from heatmaps and volcano plots responsible for the metabolic differentiation were further investigated using a pathway enrichment analysis. Modulated biochemical pathways in the distinction of ccRCCs versus pRCCs, ccRCCs versus chRCCs, and pRCCs versus chRCCs are depicted in Fig. 2.

Metabolomic differences between ROs and chRCCs utilising hierarchical clustering and k-means analysis
Using high-resolution MALDI-FT-ICR MSI, we previously identified metabolomic signatures, based on the top 50 differentially intense m/z values, which accurately distinguished ROs from chRCCs [17]. Herein, a MALDI-TOF MSI analysis of the first set yielded similar results with one chRCC misclassified (one out of 16; 6.25%; Fig. 3A), while two ROs and eight chRCCs were misclassified upon validation (ten out of 117; 8.54%; Fig. 3B).
Following an alternative approach to investigate the predictability of ROs versus chRCCs, we performed k-means analysis on the validation set utilising SCiLS Lab software. Metabolites were separated into two and up to ten clusters in an effort to identify the combination of clusters yielding the highest predictive power using R-package CARRoT [18]; this is a predictive software tool that performs model selection as per the best subset regression by using the "one in ten" rule.
The cluster signal and its percentage values were used as predictive variables with CARRoT run on each of the splits into a separate cluster. In each of the nine scenarios (ie, two to ten clusters), we set the number of cross-validations to 1000, by dividing the dataset into training (90% of the data) and testing (10% of the data) sets. For each cluster separation, we utilised the same 1000 partitions in order to facilitate the comparison of the predictive power. The

Discussion
In this work, we have provided an in situ metabolomic analysis of renal tumours previously analysed by molecular imaging. This approach expands the spectrum of 99m Tcsestamibi SPECT/CT-positive renal tumours encompassing ROs, HOCTs, LOTs, and chRCCs, and supports combined diagnostics utilising molecular imaging and histometabolomic profiling. The current study further substantiates the value of metabolomics with regard to the molecular profiling of renal neoplasms [19]. Our findings also support the feasibility of a metabolomic profiling in the subclassification of renal tumours that is currently based on a morphology-based diagnostics. In situ metabolomics provides a promising tool particularly in assessing renal oncocytic neoplasms (eg, RO vs chRCC differential) and when potentially assessing limited core biopsy specimens. According to a systematic review and meta-analysis [20], core biopsy may be unreliable in establishing a definitive diagnosis of oncocytoma. Challenging cases within the oncocytic spectrum of renal neoplasia are in fact encountered frequently, including an intermediate diagnostic category [21], cases exhibiting low-/high-grade oncocytic morphology [15,22], and other less common eosinophilic tumours, including an epithelioid angiomyolipoma, SDH-deficient RCC [23], and FH-deficient RCC [24].
Overlapping and/or misclassified cases in 2-D score plot (s) and heatmap(s) also highlight the need to integrate this metabolomic data into a proper pathological context. This is consistent with previous studies demonstrating either misclassified ccRCCs as chRCCs [16] or overlapping chRCCs and oncocytomas [25]. Although the former was suggested as a potential erroneous pathological interpretation [16], this was not the case in our series, possibly reflecting metabolomic similarities. With regard to the latter, we observed one positive chRCC within the RO subgroup (Fig. 3A) and two ROs within the chRCC subgroup (Fig. 3B). This likely reflects the limitations of the current classification and the expanding spectrum of renal oncocytic neoplasia [21,22].
As a matter of fact, two positive cases initially considered chRCCs (eosinophilic type) clustered together and separately from the negative classic chRCCs; both were reclassified as LOTs upon expert review (Fig. 1B and 1C). This newly proposed emerging renal entity is characterised by consistent morphological traits that overlap ROs and chRCCs, CK7 pos./c-Kit neg. immunoprofile, absence of multiple chromosomal losses and gains, and indolent clinical behaviour [15]. Three BHD-associated HOCT cases also exhibited a distinct metabolomic profile (Fig. 1A), further reinforcing the concept that HOCTs may represent a unique renal entity [14] and not a chRCC subtype/variant, according to the current WHO classification [26]. Whether this discriminatory metabolomic signature could directly be attributed to a specific genetic make-up associated with germline FLCN mutations remains to be investigated further [27].
The current study expands the ontogeny considerations based on the evidence emerging from molecular genetic studies. We confirmed the recent findings supporting distinctive metabolomic signatures in histogenetically related oncocytic tumours and RCC subtypes [28]. Priolo et al [25] utilised MS-based metabolomics, and demonstrated both similarities and differences between chRCCs and ROs with a clear separation in a principal component analysis scatterplot of the log 2 ratio of metabolite levels in revealing separation of RO and chRCC cases as well as volcano plots (right) exhibiting differential m/z values in RO versus chRCC cases in the (A) 99m Tcsestamibi SPECT/CT-examined set and (B) validation set. With regard to volcano plots, both fold changes (FCs, x axis; threshold 2) and p values (y axis; threshold 0.01) are log-transformed. Red dots represent significant metabolites. The further the red dot from the (0.0), the more significant the feature in the distinction of RO versus chRCC. (C) Mass spectra and ion distribution maps based on ten clusters of metabolites as generated by k-means analysis utilising SCiLS Lab software; red and green arrows indicate chRCCs and ROs of the validation cohort, respectively (left). ROC curves corresponding to the best predictive model based on separating the metabolites into two (blue), four (red), six (green), eight (yellow), and ten (black) clusters. Note a general trend of increasing predictive power with the number of clusters, and hence the best predictive power is exhibited by the model of ten clusters (right). AUC = area under the curve; chRCC = chromophobe renal cell carcinoma; CT = computed tomography; RCC = renal cell carcinoma; RO = renal oncocytoma; SPECT = single photon emission computed tomography; sPLSDA = sparse partial least-squares discriminant analysis. tumours to matched normal samples. This has also been corroborated by an untargeted in situ metabolomic approach based on MALDI-FT-ICR MSI, which clearly displayed a metabolomic distinction between ROs and chRCCs [28]. Schaeffeler et al [16] provided evidence that tumours originating from proximal nephron could be differentiated from distal nephron-derived tumours utilising targeted metabolomic/lipidomic analyses, while Steurer et al [29] identified subtype-specific metabolomic differences in proximal-derived tumours on MALDI-TOF MSI.
Our study has several limitations as the number of cases investigated using both 99m Tc-sestamibi SPECT/CT and MALDI-MSI was rather small with variable representation of tumour entities. To exemplify, neither a pRCC type II nor an HOCT was included outside of the BHD context (ie, sporadic or associated with renal oncocytosis).
Another consideration refers to the preselected spectrum of MS peaks and annotation of m/z species, that is, the range of metabolites, captured by MALDI-TOF MSI analysis. In fact, only a small number of biomarkers are actually shared between different analytical platforms and biological specimens [30]. Towards metabolic differentiation between chRCCs and ROs, urinary metabolomics highlighted citrate, carnitine, transaconitate, succinate, and p-methylhistidine [31], while tissue metabolomics revealed glycolysis and pentose phosphate pathway intermediates along with certain gamma-glutamyl amino acids as differential metabolites [25].
Hence, a prospective study is warranted at a large scale encompassing sporadic and hereditary forms of renal neoplasia, with a distribution reflective of their prevalence in clinical practice. That being said, specific in situ metabolomic signatures per tumour type shall be validated subsequently in an independent set and compared with current histopathological assessment of renal core biopsies. If not superior, complementary aspects of in situ metabolomics to current practice with regard to renal oncocytic tumours positive on 99m Tc-sestamibi SPECT/CT examination should be explored and further specified. As intratumour genetic and metabolic heterogeneity has been demonstrated in renal cancer [32], in keeping with the phenotypic intratumour heterogeneity utilising MSI [33], further studies are additionally required to investigate whether heterogeneity might confound this combined method.

Conclusions
The present study provides novel molecular insights into renal neoplasia and supports the feasibility of an integrated in situ metabolomic profiling for the diagnostics and classification of renal tumours. The results of this study suggest that renal tumours positive on 99m Tc-estamibi SPECT/CT should be biopsied and analysed in an integrated fashion to inform clinical management. This approach establishes a foundation for future studies to define more accurately in situ metabolomic signatures of various renal tumour histologies and genotypes.