State-of-the-art of FAPI-PET Imaging: A Systematic Review and Meta-Analysis

Introduction Fibroblast Activation Protein-α (FAPα) is overexpressed on cancer-associated broblasts in approximately 90% of epithelial neoplasms, representing an appealing target for therapeutic and molecular imaging applications. [ 68 Ga]Ga-labelled radiopharmaceuticals-FAP-inhibitors (FAPI) - have been developed for PET. We systematically reviewed and meta-analysed published literature to provide an overview of its clinical role. Materials and Methods The search, limited to January 1 st , 2018 - March 31 st 2021, was performed on MedLine and Embase databases using all the possible combinations of terms "FAP", FAPI", "PET/CT", "positron emission tomography", "broblast", “cancer-associated broblasts”, “CAF”, “molecular imaging”, and “broblast imaging”. Study quality was assessed using the QUADAS-2 criteria. Patient-based and lesion-based pooled sensitivities/specicities of FAPI PET were computed using a random-effects model directly from the STATA “metaprop” command. Between-study statistical heterogeneity was tested (I 2 -statistics). Results Twenty-three studies were selected for systematic review. Investigations on staging or restaging head and neck cancer (n=2, 29 patients), abdominal malignancies (n=6, 171 patients), various cancers (n=2, 143 patients), and radiation treatment planning (n=4, 56 patients) were included in the metanalysis. On patient-based analysis, pooled sensitivity was 0.99 (95% CI 0.97-1.00) with negligible heterogeneity; pooled specicity was 0.87 (95% CI 0.62-1.00), with negligible heterogeneity. On lesion-based analysis, sensitivity and specicity had high heterogeneity (I 2 =88.56% and I 2 =97.20%, respectively). Pooled sensitivity for the primary tumour was 1.00 (95% CI 0.98-1.00) with negligible heterogeneity. Pooled sensitivity/specicity of nodal metastases had high heterogeneity (I 2 =89.18% and I 2 =95.74%, respectively). Pooled sensitivity in distant metastases was good (0.93 with 95% CI 0.88-0.97) with negligible heterogeneity. metastases. non-oncological for its’ the STATA “metaprop” command [21]. Freeman-Tukey double arcsine transformation was performed to stabilise variances before pooling [21]. Between-study statistical heterogeneity was tested to assess data consistency (the higher the inconsistency, the larger uncertainty in meta-analysis results) using I2 and Cochran’s Q homogeneity test. We scored heterogeneity as low, moderate, and high. Heterogeneity may be biased by several factors [22, 23], and no recommendation exists on which value is to go with the analysis. We xed as acceptable a low/moderate level of heterogeneity (i.e. I < 75%) [23]. In case of high heterogeneity between studies, other options for data analysis (e.g. were et al. [23]. Per lesion, Egger assess funnel A p-value ≤ 0.05 Statistical STATA (STATA version StataCorp USA).


Introduction
The tumour microenvironment (TME) is a complex and dynamic framework that plays a crucial role in malignant cells' survival, proliferation, spread, and drug resistance through pro-tumorigenic signalling pathways [1,2]. This evidence has led to re-focus research and drug development that shifted from the "tumour" to TME elements, which gained interest for potential therapeutic and molecular imaging applications [3,4].
Among others, cancer-associated broblasts (CAFs) have emerged as appealing TME targets. CAFs constitute an extremely heterogeneous and plastic cell population, characterised by different origins, functions, and surface markers [4][5][6]. In particular, Fibroblast Activation Protein α (FAPα) -a dipeptidyl peptidase -is overexpressed on CAFs' cell membrane and stroma in approximately 90% of epithelial neoplasms [7,8]. FAPα is also a marker of wound healing and other active extracellular matrix remodelling processes, including liver cirrhosis and myocardial infarction [9][10][11]. In cancer pathogenesis, FAP is present on functionally crucial TME stromal cells, contributes to CAFs' tumorigenic effect, and might be associated directly with the malignant phenotype of transformed cells [12,13]. Additionally, tumour cells in osteosarcoma, glioblastoma and other neoplasms express FAPα [14,15].
Therefore, FAPα appears to be a suitable target both for oncological and non-oncological imaging. [ 68 Ga]Ga-labelled radiopharmaceuticals -FAP-inhibitors (FAPI) -have been developed for in-vivo positron emission tomography/computed tomography (PET/CT) or PET/magnetic resonance imaging (MRI).
The clinical and scienti c interest in [ 68 Ga]Ga-FAPI imaging has shown an explosive increase, as shown by the number of publications and the number of active trials (Fig. 1).
Indeed, in recent years, [ 68 Ga]Ga-FAPI imaging has been explored for various purposes in different clinical settings with promising results. The present work aimed to systematically review and meta-analyse published literature on [ 68 Ga]Ga-FAPI imaging to provide evidence-based indications on the potential role of these tracers.

Literature search and study selection
Once conceptualised, the project has been registered in PROSPERO (https://www.crd.york.ac.uk/prospero/) (registration number CRD42020222886). The systematic review was carried out following the PRISMA statement (the checklist is available as Supplementary material). A four-step search and evaluation strategy was adopted and executed independently by two reviewers (MS and FF). The rst step consisted of identifying sentinel studies within the PubMed database by applying multiple combinations of the following keywords: [ 68 Ga]Ga-FAPI, PET, cancer-associated broblasts. In the second step, speci c keywords and MeSH terms were de ned, as follows: "FAP", FAPI", "PET/CT", "positron emission tomography", " broblast", "cancer-associated broblasts", "CAF", "molecular imaging", and " broblast imaging". In the third step, the MedLine and Embase databases were searched with all the possible combinations of these terms and the resulting lists of matching manuscripts were exported in .csv format. The search was limited to the January 1st, 2018 -March 31st 2021 period. The application of a starting date was related to the rst publication on radiopharmaceutical in 2018 [18]. In the last step, the lists were fused and screened to identify papers describing the use of [ 68 Ga]Ga-FAPI PET in humans.
For article selection, the list was rst screened for duplicates, which were removed. Then, the list was screened to identify speci c keywords that identi ed papers outside the scope of the present review, such as "animal", "preclinical", "phantom", "osteomalacia", and "brown fat". These terms were used to highlight papers potentially out of the scope of the analysis. Then, the title and the abstracts of these studies were screened to con rm the exclusion.
Subsequently, the following exclusion criteria were de ned: a) full-text not in the English language; b) out of the scope of the present review and meta-analysis; c) preclinical studies without translational aspects (i.e., not involving human subjects); d) phantom, analytical, or simulation studies; e) single-patient case report; f) editorials, commentaries, and reviews, g) conference proceedings. Titles and abstracts of the identi ed articles were reviewed, applying the exclusion criteria mentioned above, and selected articles were retrieved in full-text. In the case of publications from the same research group/institution that presented signi cant overlap in terms of aim(s) and population, the study with the largest cohort was included. A reference list of selected articles retrieved in full-text was screened for potentially eligible studies. Additionally, the reference list of case reports, editorials, commentaries, and reviews was screened.

Quality assessment
The quality of each study was assessed independently by two reviewers (MS and FF) using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) criteria [19]. As per the QUADAS-2 scoring design, for the "patient selection", "index test", and "reference standard" domains, the risk of bias and applicability were evaluated. Whereas in the " ow and timing" domain, only the risk of bias was assessed.
Based on the signalling questions and the evaluation of the match of the considered paper with the review purpose, we de ned both the "risk of bias" and the "applicability" as "unclear", "low", or "high". We assigned 0.5 points in case of an "unclear" score, 1 point in case of "high risk of bias/low applicability", and "zero" in case of "low risk of bias/high applicability". Studies were excluded if they totalled 4 points or more across the seven QUADAS-2 sub-domains. A third reviewer (MK) assessed the paper blinded to previous assessments in case of discordancy, and majority voting was used for the nal decision.

Data collection
For each study, we collected the following information: 1) general features (name of the authors, year of publication, journal, country, study design, sample size, funding, con ict of interest), 2) study broad category (oncology, cardiology, immunology) and sub-category (e.g., heart remodelling, GI malignancies), 3) imaging technical aspects (patient preparation, acquisition modality and protocols, injected activity, uptake time, FAPI molecule, radiopharmaceutical/imaging modality used as comparator if any, interpretation criteria) and 4) type of image analysis (qualitative, semi-quantitative, or quantitative); 5) reference standard (pathological, morphological, functional, hybrid), and 6) nally, we collected metrics used to assess [ 68 Ga]Ga-FAPI imaging performance. Detection rate, sensitivity, speci city, and accuracy were recorded or calculated whenever possible (i.e. available number of true/false positive and true/false-negative cases according to reference standard) at a per patient and per lesion level. For studies not strictly dealing with diagnosis, we collected metrics used to evaluate the diagnostic performance of [ 68 Ga]Ga-FAPI imaging according to the speci c aim. The corresponding author of the studies was contacted in case of missing data. Data were cross-checked, and any discrepancy was discussed to reach a consensus (MS, FF, and MK).

Statistical analysis
Descriptive statistics and frequency tables were used to summarise data. We classi ed the papers according to the topic in oncological and non-oncological and then performed the analysis. Because of the objective of the present study, which was to provide evidence-based data on [ 68 Ga]Ga-FAPI-PET imaging, we included in the meta-analysis only those studies (at least three per sub-category) that provided sensitivity and/or speci city or complete data to construct a confusion matrix. Sensitivity, speci city, and their 95% con dence intervals (CIs) were calculated from each study. The upper con dence interval was cropped to 1 [20]. Forest plots of the estimated pooled sensitivities and speci cities (with 95% con dence intervals) were created. The weight of each study was calculated from the random-effects model directly from the STATA "metaprop" command [21]. Freeman-Tukey double arcsine transformation was performed to stabilise variances before pooling [21]. Between-study statistical heterogeneity was tested to assess data consistency (the higher the inconsistency, the larger uncertainty in meta-analysis results) using I2 and Cochran's Q homogeneity test. We scored heterogeneity as low, moderate, and high. Heterogeneity may be biased by several factors [22,23], and no recommendation exists on which value is adequate to go further with the analysis. We xed as acceptable a low/moderate level of heterogeneity (i.e. I 2 < 75%) [23]. In case of high heterogeneity between studies, other options for data analysis (e.g. sub-groups metaanalysis) were preferred as recommended by Higgins et al. [23]. Per lesion, analysis was further strati ed according to the type and/or the disease site (e.g. primary tumour, nodal involvement, and/or distant metastases). Publication and other potential bias were assessed using funnel plots. The Egger method was applied to assess funnel plot asymmetry. A p-value ≤ 0.05 was considered statistically signi cant. Statistical analyses were performed using STATA (STATA version 16.1 StataCorp LP, College Station, TX, USA).

Study selection
The search of the PubMed/MEDLINE and EMBASE databases returned a total of 1278 studies. Duplicates' removal eliminated 322 papers. The screening of titles and abstracts applying the criteria mentioned above resulted in selecting 36 papers, which were retrieved in full-text. Two articles, not ful lling the selection criteria (one case report and one with overlapping population), were excluded after reviewing the full text.
Thirty-four articles were nally assessed for quality ( Supplementary Fig. 1), and 23/34 (70%) were assessed as having an acceptable QUADAS-2 score (< 4). Figure 2 details the selection process. Supplementary Table 1 summarises the main characteristics of articles assessed for quality and included in the systematic review.

Systematic review
Twenty-three articles were included in the systematic review analysis. The main issues related to the quality of these studies (Fig. 3) were related to i) patient selection (39% and 43% of papers scored as having a high risk of bias and serious concerns about applicability, respectively) and ii) reference standard (30% of papers scored as having a high risk of bias). None of the studies, not even the prospective ones, reported power or sample size justi cation. One study was designed as a phase I investigation [24]. In 12/23 studies [ 68 Ga]Ga-FAPI was offered as a compassionate drug according to the German Medicinal Product Act § 13(2b) [25,26,35,36,[27][28][29][30][31][32][33][34]. In the remaining 10/23 papers, the trial phase was not speci ed or designated as "not applicable" [37][38][39][40][41][42][43][44][45][46]. In 18/23 studies [ 68 Ga]Ga-FAPI imaging was compared to other imaging technique(s). Sixteen out of 23 studies (70%) were nanced by non-pro t organisations, while none of the included studies received funding from industry or private entities. Authors declared a con ict of interest in 14/23 articles (10 related and four unrelated to work, respectively). Con ict of interest related to the work consisted of a patent application for quinoline based FAP-targeting in 9/10 cases.

Non-oncological studies
Non-oncological studies included six papers (Table 1). Three out of 6 papers were focused on cardiovascular conditions (287 patients), one on systemic sclerosis (21 patients), and the remaining two evaluated patients with IgG-4-related disease (53 patients).

Oncological studies
Seventeen studies described the use of FAPI in oncology (Table 2). Ten out of 17 articles focused on tumour staging and/or restaging: in head and neck cancer (2/10 papers, 29 patients), abdominal malignancies (6/10 papers, 171 patients), and a variety of cancers (2/10 papers, 143 patients). Four out of 17 papers were focused on radiation treatment planning (56 patients), and the remaining 3/17 dealt with biodistribution and kinetics (90 patients).

Meta-analysis
We excluded from the meta-analysis articles not focused on oncology and the three studies on biodistribution and kinetics. Finally, 392 patients in 14 studies were included in quantitative analysis ( Table 2). Papers were included in the sub-group analysis according to data availability, as detailed in Supplementary  (Fig. 4).

Lesion-based performance analysis
Estimated pooled lesion-based sensitivity and speci city of [ 68 Ga]Ga-FAPI imaging were not reliable on a per lesion level (data not shown) since they were affected by high heterogeneity (I 2 = 88.56% p = 0.001 and I 2 = 97.20% p = 0.001, respectively). Funnel plots ( Supplementary Fig. 2 Therefore, we performed separated sub-group analyses to evaluate [ 68 Ga]Ga-FAPI imaging ability to identify the primary tumour and detect nodal involvement and/or distant metastases. Estimated pooled sensitivity for the diagnosis of the primary tumour was excellent, reaching a value of 1.00 (95% CI 0.98-1.00) without heterogeneity among studies (I 2 = 0.00%, p = 0.83) (Fig. 5). Estimated pooled sensitivity and speci city of [ 68 Ga]Ga-FAPI imaging to identify nonprimary tumour (nodal and distant metastases) lesions (data not shown) were biased by high heterogeneity (I 2 = 92.66% p = 0.001 and I 2 = 95.20% p = 0.001, respectively). In particular, for nodal involvement, heterogeneity for sensitivity and speci city were as follows: I 2 = 89.18% p = 0.001 and I 2 = 95.74% p = 0.001, respectively. Funnel plots (Supplementary Fig. 3) showed an asymmetrical distribution of dots suggesting data bias which emerged statistically signi cant when testing the funnel plot asymmetry test for the nodal status speci city analysis (bias − 3.89, SE 0.803, p < 0.0001). Estimated pooled sensitivity of [ 68 Ga]Ga-FAPI imaging in distant tumour metastases detection resulted good (0.93 with 95% CI 0.88-0.97) without heterogeneity among studies (I 2 = 0.00% p = 0.41) as shown in Fig. 6.
The estimated pooled sensitivity/speci city improved when restricting the analysis to papers focused on abdominal malignancies. In parallel, we assisted in a heterogeneity reduction. Speci cally, estimated pooled lesion-based sensitivity and speci city on a per lesion level resulted 0.96 (95% CI 0.90-1.00) and 0.79 (95% CI 0.62-0.93) respectively, with moderate to low heterogeneity (I 2 = 68.05% p = 0.01 and I 2 = 18.20% p = 0.30, respectively) as shown in Supplementary   Fig. 4. Estimated pooled sensitivity for the primary tumour diagnosis substantially con rmed the ndings obtained on all studies (1.00 with 95% CI 0.98-1.00, I 2 = 0.00 and %, p = 0.95) ( Supplementary Fig. 5). Estimated pooled sensitivity for diagnosis of non-primary tumour resulted high (0.87 with 95% CI 0.82-0.92, I 2 = 22.99 and %, p = 0.27) (Supplementary Fig. 6). limitations are thoroughly studied and include: i) cancers that are well-or moderately-differentiated and, thus, present a relatively slow growth and a limited Warburg effect; ii) tumours located close to structures/organs with variable physiological/in ammatory/drug-induced uptake, such as liver and gut neoplasms; iii) tumours in areas with permanently elevated uptake, such as brain and urinary tract malignancies. plays a crucial role in tumour invasion, metastasis, and angiogenesis, and its expression is associated with several factors, including higher local tumour invasion, increased risk of nodal metastases and poorer outcome [4,56]. Therefore, the high variability of [ 68 Ga]Ga-FAPI performance in nodal staging assessment (sensitivity 59%-100%) could appear unexpected, especially considering that lymph nodes are typically structured by a network constituted by broblast reticular cells. However, the exact role (s) exerted by FAP and FAP -positive cells in cancer is still to be de ned. Evidence suggests a contextdependent functioning and that it is at least in part tumour type-speci c [13]. Some recent data on breast cancer supported these hypotheses. Primary breast tumours were found to exhibit higher FAPα mRNA levels than nodal metastases [57], suggesting the existence of distinct transcriptomic programs in broblasts located in different tissues, possibly dictated by tissue-speci c environmental cues [58].

Discussion
Recent data showed that healthy and metastatic lymph nodes are enriched by speci c CAFs populations [59]. Notably, two sub-populations (CAF-S1 and CAF-S4) are predominant in lymph nodes invaded by breast tumour. CAF-S1 overexpressed FAPα, while CAF-S4 are characterised by a low to moderate expression of FAPα with signi cant impact on outcome: CAF-S1 initiates the rst steps of epithelial-to-mesenchymal transition and secrete attractive factors for cancer cells. At the same time, CAF-S4 promotes matrix remodelling and cancer cell invasion [59]. Moreover, Ser ing et al. [33] suggested a correlation between FAPα lymph node metastases expression and lesion size (weak FAPα expression in lesion < 7 mm, which resulted negative at imaging  [25]; it could serve as a biomarker, working in synergy with the well-established cardiovascular prognostic risk factors [29], and it might have a role to assess treatment-related cardiotoxicity [35]. We found that, in approximately 2/3 of the considered studies, at least one co-author declared a con ict of interest, re ecting the growing interest of the pharmaceutical industry in theragnostics. Indeed, the demonstration of the e cacy of [ 177 Lu]Lu-DOTATATE in neuroendocrine neoplasms [90] and of [ 177 Lu]Lu-PSMA-617 in prostate cancer [91] attracted considerable investments in the radiopharmaceutical eld [8]. Sponsored trials on FAPα-targeted applications are already ongoing (Fig. 1), and more are expected shortly. Until a few years ago, studies in diagnostic and therapeutic Nuclear Medicine were generally investigator-initiated trials (IIT) rather than industry-sponsored trials (IST) [89]. Consequently, con ict of interest was a much less relevant issue, and we foresee a new research environment soon. A closer industry-academia collaboration may optimise the resources, increase the quality of the studies and ensure the safety of novel radiopharmaceuticals. Both can contribute to producing high-level evidence and to establishing new recommendations and guidelines. Awareness of the industry interest will enhance the critical appraisal of the investigations. Studies on FAPI-targeted applications are expected to signi cantly in uence clinical practice in the near future [8].
This meta-analysis presents some limitations. Firstly, the relatively small number of published articles in the eld is a possible source of bias. Secondly, the sample size and the study design of the studies included in the analysis vastly differed, possibly affecting the reliability of results and preventing the possibility of including all the studies in all sub-group analysis. Thirdly, the study design of papers included in the meta-analysis prevented the estimated pool speci city calculation for both primary tumour and metastases (Tables 1 and 2). Although these aspects may have in uenced our results and/or affected statistical power, informative data emerged.  Paper selection process.

Figure 3
Quality assessment according to QUADAS-2 of the 23 articles included in the systematic review.   Estimated pooled sensitivity for distant metastases detection.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryMaterialFAPImetaanalysis14.05.2021.docx