Spatially fractionated radiation therapy: a critical review on current status of clinical and preclinical studies and knowledge gaps

Spatially fractionated radiation therapy (SFRT) is a therapeutic approach with the potential to disrupt the classical paradigms of conventional radiation therapy. The high spatial dose modulation in SFRT activates distinct radiobiological mechanisms which lead to a remarkable increase in normal tissue tolerances. Several decades of clinical use and numerous preclinical experiments suggest that SFRT has the potential to increase the therapeutic index, especially in bulky and radioresistant tumors. To unleash the full potential of SFRT a deeper understanding of the underlying biology and its relationship with the complex dosimetry of SFRT is needed. This review provides a critical analysis of the field, discussing not only the main clinical and preclinical findings but also analyzing the main knowledge gaps in a holistic way.


Introduction: radiation spatial fractionation broadens new horizons in radiotherapy
Over the last decades technological advancements, especially in medical imaging and accelerator domains, have led to significant improvement in tumor dose conformality (Bernier et al 2004) and a notable increase of the therapeutic index for many tumor types.However, normal tissue tolerances remain the main limitation for an efficient treatment of radioresistant tumors, tumors close to delicate structures such as spinal cord or some pediatric cancers, in current clinical radiotherapy (hereafter referred as conventional radiotherapy (RT)), including the state-of-the-art treatment technologies (Hoeller et al 2021).This compromises an effective treatment of many late-stage tumors, radio-resistant bulky tumors (Kersting et al 2023), recurring tumors (Mahvi et al 2018), large brain tumors (Minniti et al 2011, Mohammadi et al 2017, Lawrie et al 2019, Singh et al 2021) and some pediatric cancers (Haddy et al 2011, Vatner et al 2018).Unable to control the tumor safely and effectively, palliative treatment is often offered in those cases, intended only for symptom management.Thereby, new therapeutic approaches which offer increased normal tissue tolerance are of utmost importance.Along this line, spatially fractionated radiation therapy (SFRT), although it is an unconventional, is a radiotherapeutic approach offering promise in those difficult-to-treat cases (Billena and Khan 2019, Yan et al 2020, Prezado 2022).SFRT holds great potential to overcome the main limitation of conventional RT, i.e. normal tissue tolerances (Billena and Khan 2019, Yan et al 2020, Prezado 2022).
The (mounting) evidence in SFRT is based on many decades of clinical use and preclinical studies (Billena andKhan 2019, Prezado 2022).This contrast with the case of some other unconventional and more recent techniques, such FLASH therapy.While a global summary of the main published results of the SFRT domain can be found in previous recent reviews (Billena and Khan 2019, Yan et al 2020, Prezado 2022), this manuscript intends to provide a critical analysis of the state of the art both from preclinical and clinical point of view, and especially the main knowledge gaps both from physics and biological angles.
The fact that SFRT intentionally uses (highly) heterogenous dose distributions represents a disruption with respect to the classical paradigm of RT, in which tumor dose homogeneity is aimed for.In SFRT the irradiation is performed with narrow beams spaced by a certain distance.The resulting dose distribution consists of areas of high dose ('peaks') followed by areas of low dose ('valleys') as represented in figure 1.A significant tumor control and reduction in normal tissue toxicities have been reported both in SFRT patients' treatments and in small animal experiments (Billena and Khan 2019, Yan et al 2020, Prezado 2022).In the latter, it should be noted that tumor control has been achieved even when highly heterogeneous dose distributions with very low valley doses were used (Prezado et al 2018, Sotiropoulos et al 2021).The remarkable normal tissue sparing and tumor control in SFRT cannot be explained with classical radiobiology concepts (Joiner and van der Kogel 2009), such as the consideration of the DNA of the cell nucleus as the main target for radiation-induced genotoxicity (Little 2003) with the goal being to induce a 'direct cell death', in each of the tumoral cells.There is increasing evidence suggesting that radiation effects in SFRT may go beyond simple cell death, and that so-called non -targeted effects, such as bystander effects (Asur et al 2015) as well as stromal and immunological changes (Lumniczky and Sáfrány 2015, Yoshimoto et al 2015, De Martino et al 2021), which take place at a more at more macroscopic scale/tissular level, may have a major role in tissue response to SFRT (Prezado 2022).Further details will be given in the next sections.
There are several forms of SFRT: GRID therapy (Mohiuddin et al 1999), Lattice therapy (LRT) (Wu et al 2020), Microbeam radiation therapy (MRT) (Slatkin et al 1992) and Minibeam radiation therapy (MBRT) (Dilmanian et al 2006).A significant number of patients have been treated with GRID and LRT (Billena andKhan 2019, Yan et al 2020).MRT and MBRT are still in the preclinical stage.The favorable preclinical data (Prezado 2022) along with the successful veterinary trials of de novo gliomas in dogs (Kundapur et al 2022) encourage the preparation of clinical trials in these two techniques.A detailed overview, including historical perspective about the different techniques can be found in the recent review by Prezado et al (Prezado 2022).
One of the main differences among the several SFRT modalities is the geometry and size of the high dose regions.See table 1, adapted from (Prezado 2022).Thus, SFRT is a highly complex approach since it is Table 1.Summary of the main features of the different techniques in SFRT.Adapted from (Prezado 2022).In the next sections, we will provide a critical discussion of the published clinical data and radiobiology experiments performed thus far.We will also describe the different technical implementations and critically discuss the existing knowledge and gaps in terms of dosimetry and biological effects.

SFRT clinical and preclinical applications
A brief and critical overview of SFRT clinical and preclinical application is provided in this section.

Discussion on SFRT clinical data
The first demonstration of the clinical advantages of SFRT dates to the beginning of 20th century (Laissue et al 2012), with GRID therapy which used a perforated screen, such that the irradiation was performed with a set of pencil beams.Nowadays, GRID therapy is delivered mainly with brass collimators attached to the head of linear accelerators (see figure 2), conventional multileaf collimator (MLC) or the binary MLC used in the TOMOGRID method (Ha et al 2006, Zhang et al 2016, Billena and Khan 2019, Grams et al 2022), which allows modulating the dose three dimensionally and improves tumor conformity.Some other alternatives for GRID generation include a method of collimation using MLCs that administers beams 2 rows at a time in a step-and-shoot fashion (Ha et al 2006) or a hybrid method using both a block and an MLC, each creating a set of parallel stripes when set perpendicular to one another to form a grid (Almendral et al 2013).A more detail discussion about the advantages or disadvantages of each method can be found in the review paper by Billena et al (Billena and Khan 2019).Finally, proton GRID therapy has been recently implemented based on proton beam scanning (PBS) technology.In this case, individual pencil beams are used directly (no collimation) (Mohiuddin et al 2020).
On the other hand, in LRT (Wu et al 2020) the spatial fractionation is only present in the target, where multiple localized high-dose spherical regions (1-2 cm size) called vertices are produced within the tumor, as represented in figure 2. Some techniques for delivering LRT include volumetric modulated arc therapy (VMAT) (Grams et al 2021), intensity-modulated radiation therapy (IMRT) (Dincer et al 2022), and proton/heavy ion beams with spot scanning nozzles (Wu et al 2020).The placement of vertices is optimized during planning to meet the dose criteria (the dose to the tumor periphery is kept to a level deemed tolerable by the surrounding normal tissues and critical organs).The planning of LRT is not standardized, and some studies used a uniform distribution of vertices within the PTV while others limit the vertices to specific regions of the tumor (Ferini et al 2022).
As modern technologies have introduced new ways of delivering SFRT and advantageous preclinical data has been accumulating, there is a renewed interest for SFRT within the Radiation Oncology community.A number of SFRT clinical trials are ongoing or in the planning stage.As of today, and to the best of our knowledge, only GRID and LRT therapy have been used in patients' treatments.Clinical data compilations can be found elsewhere (Billena and Khan 2019, Yan et al 2020, Moghaddasi et al 2022, Iori et al 2023).Hereafter, we will critically discuss that data.
The literature reports treatment of approximately 500 patients with SFRT, mainly with advanced disease and palliative intent (Mohiuddin et al 1990, Mohiuddin et al 1996, Mohiuddin et al 1999, Huhn et al 2006, Peñagarícano et al 2010, Neuner et al 2012, Amendola et al 2019, Choi et al 2019, Amendola et al 2020, Mohiuddin et al 2020, Pollack et al 2020, Snider et al 2020, Duriseti et al 2022, Ferini et al 2022, Ahmed et al 2023).Among them, 69% were treated with LRT and 31% with GRID.It should be noted that due to the beam widths (~1 cm) and spacing (several cms) in GRID and LRT, those treatments are mostly limited to bulky tumors.All reports are retrospective institutional series except two (Pollack et al 2020, Duriseti et al 2022).
Palliative GRID or LRT were used in various tumor types and locations and mostly before or after conventional external beam RT.They were delivered as a single or limited-number of high-dose fractions using photons (Mohiuddin et al 1990, Mohiuddin et al 1996, Mohiuddin et al 1999, Neuner et al 2012, Amendola et al 2019, Choi et al 2019, Duriseti et al 2022, Ferini et al 2022, Ahmed et al 2023), except for one study that employed protons (Mohiuddin et al 2020).Globally, in the studies on palliative SFRT, the percentage of clinical improvement of symptoms ranges from 54.5% to 100% with median follow-ups of less than one year.SFRT with palliative intent was reported to be well-tolerated.The percentage of grade 3 or higher adverse events (including bleeding (head and neck target), mucositis, small bowel obstruction, and grade 5 tumor lysis syndrome (Schiff et al 2022)) are reported to range from 0% to 24% (Mohiuddin et al 1990, Mohiuddin et al 1999, Neuner et al 2012, Amendola et al 2019, Choi et al 2019, Mohiuddin et al 2020, Ferini et al 2021, Duriseti et al 2022).A recent published study (Ahmed et al 2023) reports on a large cohort of 53 sarcoma patients treated with SFRT and followed by a conventional radiation therapy with palliative intent in more than 80% of the cases.Symptom relief was documented with 32 treatment courses (60%).One-year overall survival and local control rates were 53% and 82%.Stable or partial response was observed with 47 treatment courses (90%).Four grade 3-4 acute and subacute toxicities were attributable to SFRT (8%).
On the other hand, SFRT with curative intent for advanced disease or with ablative purpose in metastatic lesions was also employed notably in squamous cell carcinomas of the head and neck, sarcomas, cervical and prostate cancer (Mohiuddin et al 1999, Huhn et  All studies in patients with curative intent and advanced disease used SFRT in combination with conventional RT; the treatment consisted of adding one session of GRID or LRT using photon beams and a prescription dose from 10 to 20 Gy, except for one protocol that used 24 Gy in three fractions for gynecologic tumors (Amendola et al 2020).One-year local control ranged 80%-100% but was not always reported.Toxicity was seen in up to 50% of the patients, especially skin and mucosal reactions with head and neck cancer.
As mentioned previously, most reports were retrospective institutional series, sometimes results were mixed with palliative patients, except two.Duriseti et al (2022), a single-arm phase-1 trial, included 20 patients with advanced or metastatic disease (>4.5 cm tumor diameter) treated with Lattice stereotactic body RT which delivered 20 Gy in 5 fractions with a 66.7 Gy simultaneous integrated Lattice boost using volumetric modulated arc therapy.Three patients were treated with curative intent as definitive RT.No grade-3 toxicity was reported.One patient treated to the abdomen had urosepsis requiring hospitalization (grade 4).Patient reported outcomes (PROMIS) demonstrated that patients had improved physical function, pain, global health, anxiety, and depression at 14 d after treatment and this improvement was preserved at 90 d; tumor volume decreased (median −24.4% by volume) following Lattice treatment (Duriseti et al 2022).A second one phase-1 trial included 25 patients with favorable to high-risk prostate cancer (Pollack et al 2020).A partial tumor volume based on multiparametric MRI received a 12-14 Gy Lattice session before a conventional treatment.With a median follow-up of 66 months, there were no grade-3 genitourinary or gastroinstestinal adverse events; two patients had biochemical failure.
The fact that the studies included very different tumors and locations, various combinations with conventional radiotherapy, used different irradiation geometries and dose prescription, make it a challenge to extrapolate solid conclusions concerning the best dose parameters, distribution pattern, etc.Moreover, the lack of prospective and comparative data to conventional treatment prevents definite conclusions as to the efficacy of SFRT over standard of care and this technique should be considered as a therapeutic option within clinical trials.

Discussion on SFRT preclinical data
There are numerous published preclinical studies showing the ability of SFRT to minimize toxicity while leading to significant tumor control.Extensive details can be found in recent reviews (Meyer et al 2017, Billena and Khan 2019, Yan et al 2020, Prezado 2022).The following is a brief critical summary of preclinical experiments on normal tissue sparing and tumor control with SFRT, and while not exhaustive, should serve to highlight the unique properties and potential of SFRT as a treatment modality.
The convenient size and spacing of the narrow (submillimeter) beams used in MRT and MBRT for rodent irradiations is the reason why most of the preclinical data was acquired with these two techniques, although a few experiments using GRID and LRT were also carried out in tumor bearing animals (Kanagavelu et al 2014, Sharma et al 2017, Johnson et al 2022).Toxicity studies were only performed in MRT and MBRT.
The extreme and unique characteristics of MRT require the use of kV beams (inadequate for deep-seated tumor treatments) (Prezado et al 2009).Compared to kV x-ray beams, the substantial lateral phantom scattering of MV beams would increase the dose in the valleys, minimizing the spatial dose modulation in MRT.The use of charged particles, e.g.protons, carbon or electrons for MRT is not favorable either (Martínez-Rovira et al 2015).Due to the unavoidable nature of Compton scattering, microbeams of charged particles would quickly blur together and eliminate much of the dose heterogeneity.The very high-dose rates needed to keep the characteristic peak-and-valley dose profiles intact when micrometric sized beams are to be used (Manchado de Sola et al 2018) has confined MRT to a few large synchrotron light sources such as the european synchrotron radiation facility (ESRF) (France) (Laissue et al 1998).Recent technical developments aim to produce more compact sources for MRT, such as carbon nanotube systems (Hadsell et al 2013) or line-focus x-ray tube (Winter et al 2020).However, currently only prototypes exist, and the techniques will be restricted to low energy x-rays.
On the other hand MBRT has been implemented at small animal irradiators (Bazyar et (Schneider et al 2020, Schneider et al 2021).The use of static collimators is a straightforward method to implement MBRT at any facility.However, it suffers from lack of flexibility (a different collimator should be manufactured for each irradiation pattern, e.g.mini beam width and center-to-center (ctc) distance), and lack of efficiency (the flux is reduced).In recent years, some alternative and more flexible methods have been proposed.Those include a scanning dynamic collimator for protons and heavier ion MBRT (Sotiropoulos and Prezado 2021) or a versatile collimator for x-ray MBRT consisting of a collection of alternating tungsten or brass plates, combined with 3Dprinted plastic plates that can be stacked together in the desired order (Stengl et al 2023), as showed in figure 2. Further details on MBRT generation can be found elsewhere (Schneider 2022).
The remarkable normal tissue sparing in SFRT has been observed in different animals models (mice (Serduc et al 2006, Serduc et al 2008a, 2008b, Brönnimann et al 2016), rats (Dilmanian et al 2002, Regnard et al 2008, Prezado et al 2015, Prezado et al 2017b), rabbits (Laissue et al 2022), etc) and organs (brain (Prezado et al 2017a, 2017b, Lamirault et al 2020b), bone (Laissue et al 2022) lung (Trappetti et al 2021), spinal cord (Jaekel et al 2023), etc).The tolerance dose level is highly influenced by the beam width according to the well-known phenomenon of dose-volume effects (the narrower the beam size is, the higher the tolerance of normal tissues is (Zeman et al 1961)).One example on how the beam width influences dose tolerance in SFRT is the work by Sammer et al (2019).The extremely narrow beams employed in MRT (25-100 μm) allows that normal tissues exposed to MRT can withstand enormous peak doses of around 600 Gy in one fraction in unilateral irradiations with 50 μm wide beams (Laissue et al 2001, Serduc et al 2009).An example of the remarkable normal tissue sparing of MRT is the historical experiment of irradiations of weanling pig cerebellums by Laissue et al (2001).They showed that that microbeams of approximately 20 micrometers width spaced 210 micrometers delivered entrance doses as high as 600 Gy to the cerebellum of young pigs and no adverse effects were observed after a follow up more than 1 year (Laissue et al 2001).Toxicities have been observed after MRT in preclinical studies, such as in brain and lung irradiations at extreme peak doses of 800 Gy (Serduc et al 2009, Bräuer-Krisch et al 2010, Trappetti et al 2021) and recent studies irradiating rabbit facial bone with MRT showed peak doses greater than 200 Gy caused severe dental damage (Laissue et al 2022).
In MBRT, where thicker beams (0.5-1 mm) are used, peak doses as high as 100 Gy were reported to be well tolerated (Deman et al 2012, Prezado et al 2015).Toxicity is observed at 150 Gy or above peak dose (depending on the beam width) (Prezado et al 2015).Remarkable normal tissue sparing of MBRT was observed with different particle types in MBRT (photons (Prezado et al 2017a), protons (Prezado et al 2017b)carbon (Dilmanian et al 2012) and Ne ions (Prezado et al 2021)).For tumor control, however, it should be noted that the smaller the beam is, the higher the dose needed to control the tumor (Regnard et al 2008).
Both MRT and MBRT lead to a widening of the therapeutic window for aggressive tumor models in preclinical experiments.While MRT needs hundreds of Gy peak dose to ablate radioresistant tumors, MBRT has shown to be able to achieve similar treatment response with 50-80 Gy peak dose, a much safer dose level for normal tissue toxicity.
Therefore, for clinical application SFRT requires a good compromise between the beam width and doses to simultaneously achieve both normal tissue sparing and tumor control.Along this line, it should also be highlighted that to advance SFRT it is crucial to employ a relevant rationale when comparing techniques.Future fair comparison among techniques can help establish what tumor types could most benefit from one technique or the other.But those comparisons should have a clinical basis and should be based on how much we widen the therapeutic window and not only on how normal tissues can withstand increasingly higher doses.Following the ALARA principle (as low as reasonably achievable), any RT technique should deliver doses high enough to control the tumor with minimal toxicity and no more than that.Thus, the comparison must be done at a therapeutic dose for each of the SFRT techniques (see table 1).In that regard studies like (Brönnimann et al 2016) reporting MBRT or GRID to be less optimal than MRT since vascular damage is observed with 500 Gy are misleading since 500 Gy is 10 times higher than the dose needed to ablate radioresistant tumors with MBRT.The same applies to some other studies in which therapeutic GRID-RT doses (15-20 Gy peak dose) were applied in MBRT settings (Johnson et al 2022), with subsequent lack of tumor control.
Concerning tumor control, most studies in MRT have shown tumor growth delay, and tumor eradication in some configurations (mainly high peak doses being used >600 Gy).A compilation of studies can be found in Fernandez-Palomo et al (2020).However, most of the studies have not included a comparison with conventional RT irradiations.In the few studies where that comparison has been done, the dose in the standard irradiation group equaled the valley dose in MRT.Thereby, the superior tumor control in MRT can be challenged as due to a much higher integral dose than in standard RT.Most of the studies targeted brain tumors with a few experiments irradiating mouse melanoma Although MRT and MBRT are still in the preclinical stage, some recent veterinary clinical trials have been carried out with excellent results (Adam et al 2022, Kundapur et al 2022).A recent publication reports the shortterm results of the first MRT treatment of a dog with spontaneous glioma.Tumor volume reduction was observed accompanied by an absence of toxicity in the 3-months short-term follow up (Adam et al 2022).Kundapur et al (2022) reported on the first trial in dogs with spontaneous gliomas.Out of the 8 dogs treated with MBRT (26 Gy), 71% exhibited completed pathological remission without toxicity, while none of the tumors irradiated with conventional RT was eradicated.The main limitation of this study is that diagnosis was performed based on the imaging study.None of the dogs underwent biopsy as per standard of practice.

Biology mechanisms in SFRT: what we know and what we do not
As was mentioned in the introduction, the biological effects in SFRT go beyond 'simple' cell death and involve two other levels: tissue/organ level, and systemic level.The full picture is incomplete, but many pieces of the puzzle have been collected in the last years.The priority should be now to fit all of the pieces together.SFRT's mechanisms of action proposed in the literature include differential vascular effects (Bouchet et

Vascular effects
Since SFRT uses very high doses, one could expect to observe a similar vascular damage as in standard radiosurgery (Kozin 2022).Indeed, high-dose (>8-10 Gy) conventional irradiation preferentially targets endothelial cells lining tumor blood vessels by activating an acid sphingomyelinase (ASMase)-mediated apoptotic pathway.The tumor response at such high doses is predominantly due to secondary cell death caused by severe vascular deterioration (Kozin 2022).This might be the case In GRID and LRT.Increased sphingomyelinase activity and ceramide levels were only observed in patients with complete or partial response to GRID therapy (Sathishkumar et al 2005).Ceramide has been linked to microvascular endothelial cell sensitization to radiation-induced apoptosis (Garcia-Barros et al 2003).Similar observations were made in an in vivo study in which Lewis lung carcinoma bearing mice were treated with LRT (Kanagavelu et al 2014).Their serum exhibited increased acid sphingomyelinase levels.
On the other hand, a differential vascular effect in the normal versus tumoral tissues has been observed in MRT (Bouchet et al 2015).MRT exhibits a preferential damaging effect on the immature vessels, while mature microvasculature is preserved (Bouchet et al 2010, Sabatasso et al 2011, Bouchet et al 2015).MRT does not modify any structural or functional parameters such as blood volume, vascular density, and perfusion in normal tissues (Bouchet et al 2010, Sabatasso et al 2011, Bouchet et al 2015).This has led to the general claim in the synchrotron MRT community that immature blood vessels in the tumor will be more sensitive to MRT while the healthy tissue mature blood vessels will be resistant to MRT (Bouchet et al 2015).Altered blood vessel integrity, decrease in the number of vessels and significantly reduced blood perfusion were reported in the tumor in studies comparing MRT with broad beam irradiations (Sabatasso et al 2011, Potez et al 2019, Sabatasso et al 2021).However, the comparisons were done with broad beam dose equaling the valley dose in MRT, thereby much lower integral dose, which may not be a fair comparison.
Some other studies contradict this claim since they observed repair and reorganization of tumor vasculature after MRT and MBRT (Sabatasso et al 2011, Griffin et al 2012, Fontanella et al 2015, Sabatasso et al 2021).In contrast, important vascular damage was observed after conventional irradiations since the dose was high (>8 Gy).The apparent reorientation of the tumor vasculature observed after MBRT in the perpendicular direction to the beams suggests a possible migration of cells in the valleys towards the peaks.This result is not contradictory with tumor control, but on the contrary, it can favor one of the key factors in the SFRT anti-tumor response which is immune infiltration (see next subsection) where a functional vasculature is needed.
No systematic and robust evaluation has been done on the effects of other SFRT techniques on the vasculature.The only published data report on normal vascular damage after MBRT at doses corresponding to the MRT domain (>300 Gy peak dose) and much higher than the therapeutic doses in MBRT (50-100 Gy) (Brönnimann et al 2016).

Radiation-induced immune response
There is increasing evidence of the determinant role of the immune system in the anti-tumoral response of SFRT.A systematic and thorough recent review on several aspects of the radiation-induced immune response of SFRT can be found in Bertho et al (Bertho et al 2023).A summary mainly focusing on primary results and differential observations with respect to conventional RT is reported in this manuscript.
Concerning the impact of SFRT on the tumor microenvironment, two recent works (Bazyar et al 2021, Bertho et al 2022b) demonstrated the crucial role of T cells in the anti-tumor response in MRT and MBRT by showing that the absence of efficient mature CD8 + T cells prevented the response to SFRT treatments unlike immunocompetent animals.This was not observed in conventionally irradiated animals (Bertho et al 2022a), irradiated at the same mean dose.In addition, an increased tumor infiltration by CD8 + T cells and B cells was observed in both evaluations after SFRT.Moreover, Bertho et al (2022b) observed that MBRT led to a faster and more efficient (48 h post-irradiation) immune tumor infiltration, dominated by CD8 + , CD4 + and double positive T cells at the center of the tumor, compared to conventionally irradiated tumors.Additionally, to observe a significant immune recruitment after conventional irradiation, high doses and hypofractionation schemas may be needed, which sometimes are not compatible with the tolerances of surrounding normal tissues contrary to SFRT.Enhanced infiltration of CD3 + T cells (no evaluation of the specific T cell subset) and CD4 + lymphocytes and natural killer (NK) cells were also reported in primary tumors (Lewis mouse lung carcinoma) irradiated with LRT (Kanagavelu et al 2014) and MRT-irradiated (B16-F10) melanoma bearing mice (Potez et al 2019), respectively.
Moreover, SFRT might offer effective combination with immunotherapy as several promising results have been observed in preclinical studies using different tumor models (Johnsrud et al 2020, Bazyar et al 2021).
It has been hypothesized (Jiang et al 2020) that the immunomodulatory effects of SFRT have their origin in the high dose regions (peaks) resulting in the release of tumor neoantigens via the induction of tumor cell death.The release of tumoral neoantigens stimulates the activity of antigen-presenting cells.These cells are then able to prime T cells against the tumor.On the other hand, immune cells present in the tumor will be preserved in the valley areas, receiving a low dose.However, this hypothesis on reprogramming the tumor microenvironment (TME) by SFRT towards a more immunogenic TME remains to be explored and validated.Further studies are necessary to fully understand the mechanisms associated with the immunomodulatory effects of SFRT.

Abscopal effects
High radiation doses may also have a systemic effect on distant sites away from the irradiated lesion, referred to as the abscopal effect (Craig et al 2021).Accumulating evidence now suggests that the immune system is a major determinant in regulating the abscopal effect (Craig et al 2021).
Abscopal effects have been observed in several SFRT experiments (Kanagavelu et al 2014, Johnsrud et al 2020, Yan et al 2020).One of the most explicit ones is the study of Kanagavelu et al (2014) in which xenograft A549 lung adenocarcinoma was implanted in the two flanks of mice.One of the two tumors was irradiated with LRT.The growth of both tumors was reduced, and the untreated tumor then also responded more than expected with 2 Gy fractionated RT.There was an increase in the number of infiltrating T cells in both the irradiated site and the distant site.This correlated with the deceleration of tumor growth.
Abscopal effects were also observed in healthy tissues.Ventura et al (2017) found an increased number of clustered DNA lesions, double-strand breaks, apoptotic cells as well as local and systemic immune responses in a variety of non-targeted healthy tissues.No significant difference was observed between MRT and conventional irradiations.
A high dose of irradiation might not be sufficient to trigger an abscopal effect: the combination with immunotherapy may help trigger a robust innate and adaptive immune response, allowing the elimination of tumor cells outside the irradiation field at the systemic level (Lugade et al 2005, Dewan et al 2009, Lee et al 2009, Lumniczky and Sáfrány 2015).Some examples of successful combinations of SFRT and immunotherapy include MRT with gene-mediated immunoprophylaxis (Smilowitz et al 2006), MRT and anti-CTLA-4 immune checkpoint inhibitor (Bazyar et al 2021) or GRID-RT with anti-CTLA-4 and anti-PD-1 immune checkpoint inhibitors (Johnsrud et al 2020).
Finally, the study of Bertho et al (2022b) provided the first evidence of a long-term anti-tumor immunity being activated after MBRT in an orthotopic glioma model.

Cell signaling and bystander effects
Accumulated evidence in the last decades indicates that beyond DNA damage, both extranuclear targets and extracellular events may play an important role in determining the biological responses to ionizing radiation (Daguenet et al 2020).Bystander ('cohort') effects have been reported in the literature after SFRT irradiation, but mainly in vitro.A significant increase of TNFα was observed in the serum of patients treated with GRID (Sathishkumar et al 2002) and an in vivo study using mice (Kanagavelu et al 2014) with allogenic Lewis lung carcinoma receiving LRT.The involvement of TNF-α in the bystander signaling pathway has been reported (Shareef et al 2007, Wang et al 2020) by different authors.An increased expression of genes involved in DNA repair, cell cycle arrest, heat shock protein (HSP) and apoptosis were observed in the cells of the valley regions after exposure of cell cultures to GRID therapy (Asur et al 2012, Asur et al 2015).Bystander effects have also been evident in MRT, both in vivo and in vitro (Lobachevsky et al 2015, Smith et al 2018).
Some positive bystander effects have been hypothesized to explain normal tissue sparing (Dilmanian et al 2007).Some experimental indirect evidence was reported including increased levels of HSP 71 turnover found in the non-irradiated brain hemisphere of rats (Smith et al 2018), which is known to protect against neurological disorders.

Stem cell migration and proliferation
Cell migration and proliferation of stem cells in the low dose regions (valleys) to repair the tissue regions under the high dose areas (peaks), was hypothesized as one of the main factors in the remarkable normal tissue capacity of SFRT (Dilmanian et al 2002).The experimental evidence is rather limited though and only three experiments provide a certain level of proof.
The notable tolerance of rat skin to MRT and the rapid regeneration of the damaged segments of skin in the work of Zhong et al (2003) was attributed to the surviving clonogenic cells in the valley regions.A more recent study by Fukunaga et al (Fukunaga 2021) showed, for the first time, that the survival and potentially migration of the non-irradiated germ cells in the irradiated testes are required for the tissue sparing effect for spermatogenesis.
Finally, in an in vitro investigation (Crosbie et al 2010), different migration patterns between tumor and healthy tissues were observed: 24 h after MRT, peak and valley-irradiated zones were indistinguishable in tumors because of extensive cell migration while normal skin cells appear to undergo a coordinated repair response.

Biochemical mechanism
The investigation of relative production of reactive oxygen species (ROS) in SFRT is of relevance due to their potential role, amongst various biological processes, such as cell signaling (Azzam et al 2012), thereby implicating them in the aforementioned mechanisms of SFRT efficacy.
The diffusion of hydrogen peroxide (H 2 O 2 ) from the peaks leading to a homogeneous H 2 O 2 distribution over the target has been proposed as a surrogate for the dose to explain the tumor response after MRT and MRBT (Dal Bello et al 2020, Zhang et al 2023).Their calculations showed that the H 2 O 2 produced in the peaks diffuses to the -valleys.In another a recent Monte Carlo study (Masilela and Prezado 2023) the relative production of different ROS in the peak and valley regions was assessed in different primary beams types, finding differences only in the case of carbon MBRT.In the latter, the valleys were subject to a higher • OH and aqueous electron yield, and lower H 2 O 2 yield than the peaks.Linked to that,the anti-tumor immune response is increased for lower levels of extracellular H 2 O 2 (Deng et al 2020).
Table 2 summarizes the main known differences between SFRT and conventional RT, as well as some main open questions.One of the difficulties when trying to determine the distinct radiobiology in SFRT is how to best compare it with conventional irradiation, both in terms of temporal fractionation and doses.In addition, many preclinical experiments, especially in MRT do not include a conventional RT arm.

SFRT: medical physics aspects
SFRT not only brings a different radiobiology, but it also requires distinct dosimetry protocols, and a different mindset when prescribing the dose and performing treatment planning.This section will critically review those important aspects in SFRT both for clinical and preclinical use.

Dosimetry in SFRT
The dosimetry protocol to be used for the different SFRT techniques is typically determined by the shape and size of the beamlets and the dose distributions they create.We may separate GRID and LRT on one hand and MRT and MBRT on the other.Dosimetry of GRID and LRT techniques is relatively straightforward as the high dose regions are typically on the order of 1 cm or more in diameter (Li et al 2023).Any quality assurance device or dosimeter with resolution sufficient to resolve those regions is appropriate for those cases.The size of these high dose regions is similar to Although the high dose regions in LRT plans are not usually small enough to pose challenges for dose calculations, these plans can be quite complex to deliver and therefore verification that planned and delivered doses agree is warranted.As all SFRT techniques involve steep dose gradients, including GRID and LRT, a dosimeter having adequate spatial resolution is required.A variety of such detectors are available including high resolution diode arrays, electronic portal imaging devices (EPIDs) (Duriseti et al 2021), and radiochromic films (Ha et al 2006).
What is lacking in current treatment planning software is the incorporation and reporting of SFRT-specific dosimetry parameters.This will be discussed in section hereafter.This is perhaps understandable, since at the present time the underlying biology behind the effectiveness of SFRT is not well understood which makes it difficult to determine what dose metrics would be clinically relevant.
Future versions of TPS software which can incorporate the unique biologic features of SFRT such as immunomodulation into dose calculations in order to accurately predict tumor response would represent a significant and welcome innovation (Asperud et al 2021).
On the other hand, the narrow beams employed in MBRT and MRT (50-1000 μm in their narrowest direction) come with several challenges and place them in the small-field dosimetry domain.Yet, the size of such beams is considerably smaller than the beamlets employed in conventional small-field RT techniques (e.g.stereotactic radiosurgery (SRS)), which are of the order of centimetres.Consequently, additional considerations from the ones stated in the small-field dosimetry codes of practice, i.e. the TRS-483 report for Dosimetry of Small Static Fields Used in External Beam Radiotherapy (IAEA 2017), are required.
The first consideration applies to the lack of lateral charged particle equilibrium (LCPE) on the beam axis for such narrow fluence profiles and high dose gradients.The availability of detectors that possess tissue equivalence and that can cope with these dosimetry characteristics is rather limited (Bräuer-Krisch et al 2010).Radiochromic films have been the gold standard for MRT and MBRT relative dose measurements, since they provide a wide dose range and a high spatial resolution when combined with adequate reading systems (Martínez-Rovira et al 2012, Peucelle et al 2015).Some commercial solid state detectors such as the PTW microdiamond detector have been also proven suitable for dose measurements both in MRT (Livingstone et al 2018) and MBRT (Guardiola et al 2020).Some other commercial solid state detectors, such as nanoRazor diode, with a spatial resolution of 60 μm, deemed to be suitable for minibeams (De Marzi et al 2018).
Despite their good performance, both types of detectors come with their own limitations.Measurements with solid-state detector are arduous and time-consuming since the dose needs to be measured at numerous points to accurately resolve the high dose gradients.Regarding film measurements, they need post-processing, not allowing real-time dose measurements.In addition, the dynamic dose range of some of these dosimeters may be not enough to measure accurately and simultaneously the high (peak) and low (valley) doses within the regions of interest in all configurations.For instance, in MRT, where doses range from nearly 0 to 300-600 Gy, two radiochromic films might be needed to be used to measure peak and valley doses independently (Sammer et al 2019).Additionally, the film reading process is subject to various source of uncertainty as reported elsewhere (Sorriaux et al 2013), amounting up to 5% to 10%.In addition, the high dose rate in MRT limits the use of other real-time high spatial resolution detectors such as MOSFET sensors since the saturation of their dose response.Systems, such as X-Tream, consisting of a single strip of silicon able to resolve microbeam widths of 50-100 μm and measure the instantaneous dose rates in MRT applications (Petasecca et al 2012).
Ideal dosimeters for MRT and MBRT should allow the measurement of dose distributions in real-time and 2D measurements.First advances towards this end are reported in the work by Flynn et al (2023), which employs a high temporal and spatial resolution CMOS detector, whose suitability to assist with quality assurance tests for proton MBRT has been already reported.Other alternatives would also be 2D arrays of high-resolution detectors, such as the microdiamond detector.
The second consideration is the high susceptibility to geometric imprecisions in collimated beams.As reported in the work by Ortiz et al (2022), small uncertainties or variations in setup parameters, not relevant in other RT techniques, may significantly affect proton MBRT dose distributions.A representative example is the reduction in the PVDR by up to 20% for a tilt angle between the beamline and pMBRT multi-slit collimator of 0.25 degrees, as it is illustrated in figure 3.
Equivalent conclusions are expected to be drawn when applying this type of sensitivity studies to other MBRT and MRT techniques due to the similar collimation methods and dose distributions.
Taking all these considerations into account, preclinical dosimetry protocols have been proposed (Prezado et al 2011, Prezado et al 2012, Ortiz et al 2022).Those are two-steps protocols and could be extended to the future clinical trials.First, the reference dosimetry is performed in a seamless irradiation field with an ionization chamber.Then, output factors, previously assessed with high-spatial resolution detectors, enable the determination of peak and valley doses.These dosimetry protocols have been successfully used for MRT and MBRT dosimetry in preclinical experiences to ensure the reproducibility of irradiations.Further developments towards standardized SFRT dosimetry protocols and codes of practice could imply an absolute dosimetry protocol in SFRT conditions, traceable to primary standards such as small-core calorimeters, as first attempted Flynn et al (2023).
MRT and MBRT dose computation requires the development of adapted dose calculation engines since the irradiation conditions differ significantly from those in conventional radiotherapy.This implies (i) adequate physics parameters to consider the effects of the lack of LCPE and (ii) high resolution scoring grids to resolve the small beamlet sizes and high dose gradients.As reported in various studies (Ortiz et al 2022, Ortiz 2022), simulation parameters that are more likely to lead to miscalculation of dose distributions in MRT and MBRT are the threshold for production of secondary particles and scoring voxels dimensions.Values for default in conventional RT for these parameters may lead to an underestimation of peak valleys and related quantities (e.g.PVDR), and volume-averaging effects within the scoring grid, respectively.Whereas the modification of those parameters is straightforward in Monte Carlo (MC) codes, currently most clinical TPS may not allow adapting physics and scoring parameters to the requirements for correct MRT and MBRT dose calculations.However, MC calculations, although being very precise when used properly, are excessively time-consuming for clinical routine.One possibility to overcome this difficulty is to use fast calculation algorithms with hybrid dose calculation algorithms employing micro and minibeam kernels as an input in TPS (Donzelli et  Micro and minibeam production and the irradiator geometries differ significantly in many cases from those in conventional RT.For instance, such submillimeter beamlets are typically created by mechanical collimators placed in the beam path, and in MRT, the beam generated from a wiggler tuned by tens of meters of beam modifiers is highly linearly polarized  unconventional beam characteristics and production methods is relatively straightforward in several MC codes, it is very limited in commercially available TPS.Only a few of them currently allow the implementation of collimators with the submillimeter apertures typically used in these techniques.Part of these difficulties were shared by the TPS of small animal irradiators.At the time of writing this manuscript, some commercial TPS were already available offering research modules able to perform accurate calculations in MBRT.

Dose prescription: SFRT needs a distinct metrics and treatment dose prescription
SFRT requires a new mindset in terms of dose prescription and planning.SFRT comes in different scales (GRID/ LRT, minibeam, and microbeam) and different forms (peak dose in planes, pencil beams, or vertices).See table 1. Different dosimetric and geometric parameters may lead to differences in treatment response, and thus, should be considered in the planning and dose prescription.Currently, two different approaches are used to assess the dose distributions of SFRT: (1) The first approach uses dosimetry parameters associated with conventional RT.These include organ specific dose-volume histograms (DVHs) or specific points on the DVH curve, e.g.lung V20Gy (lung dose tolerance) and minimum dose to the planning target volume (PTV).
(2) The second one employs distinct dosimetry parameters that specifically characterize the spatial distribution of the delivered SFRT dose.These include peak width, valley width, peak dose, valley dose, and peak dose to valley dose ratio, etc.
Regarding dose prescription, the most frequently used method, both clinically and preclinically and, with the exception of LRT, is to prescribe in terms of peak dose at the entrance or at the depth of maximum dose deposition (Billena and Khan 2019).This method is based on historical reasons coming from the beginning of GRID and the use of orthovoltage machines.However, and despite the lack of solid scientific reasons, it continues to be recommended in recent published guidelines (Mayr et al 2022).The latter also recommend the use of equivalent uniform dose (EUD) (Niemierko 1997).However, EUD models currently used are based on the linear-quadratic model, which assumes radiation-induced clonogenic cell death is only affected by the radiation dose the cell receives.None of the bystander effect, abscopal effect, vascular effect, and immune modulation that have been considered unique aspects of SFRT irradiations is modeled in the EUD model and thus it may not be suitable for SFRT application (Guardiola et al 2018).Alternative radiobiological models considering the above are needed.Whether a modified EUD model would be enough to describe the complexity of SFRT is yet to be experimentally determined.
A step forward is a recent paper on radio-immune response modeling for SFRT by Cho et al (2023), in which a mathematical model of immune response during and after radiation for Immune response of host body and immune suppression of tumor cells has been used.The model suggests that SFRT can make a significant difference in tumor cell killing compared to the homogeneous dose distribution.SFRT might increase or moderate the cell killing depending on the immune response triggered by many factors such as dose prescription parameters, tumor volume at the time of treatment and tumor characteristics.
Certainly, widespread and potential mainstream use of SFRT would be benefit from the identification of dosimetry parameter(s) in SFRT which correlated directly with treatment response or organ toxicity, the development of novel and efficient multi-objective optimization delivering a SFRT pareto front or the conceptualization of a new metric able to encapsulate the multiparametric and multiscale nature of SFRT.Some attempts have been made in that last direction.Anderson et al (2012) proposed to use a new quantity called percentage volume below 10% (percentage volume of normal tissue, within the volume traversed by the microbeam array path, receiving a dose below 10% of the peak entrance dose) to characterize normal tissue dose distributions.However, no experimental evidence was provided.Lansonneur et al (2020) proposed the concept of dose prominence to overcome the challenge of evaluating and reporting the PVDR in a patient due to the marked inhomogeneity of the 3D dose distribution.This concept, defined here as the dose difference between a peak and its lowest contour line, is extensively used in topography, where it measures how much a peak stands out from the surrounding signal baseline.However, none of those new proposals address all the existing challenges regarding dose prescription in SFRT.
In this context there is an urgent need to both increase the number and prioritize the investigations aimed at either finding the best correlation between the individual dosimetry parameters and the biological response in SFRT or defining a new metric and experimentally validating it.One way of shedding some light in that direction, would be to perform retrospective data analysis.In this regard and concerning current existing published clinical data (section 2.1), the task is challenging.This is primarily a consequence that SFRT is often not used as a monotherapy, but as a priming therapy followed by a conventional course of RT, or is temporally interlaced with conventional RT (Duriseti et al 2022).Moreover, most of the clinical trials are monocentric with no real consensus or standardization among different centers.Additionally, most clinical studies have been prescribed in peak dose at the entrance or at the D max , implying that tumors having very different volumes or depth will get very different valleys doses or PVDR.A recent attempt to standardize the dose prescriptions has been made (Zhang et al 2020) but the recommendation continues being to prescribe in peak dose, which as previously discussed has a historical rather than scientific rationale.On the other hand, preclinical studies, whose design is more flexible as it is not limited by strict clinical constraints, are better suited to shed light on the aforementioned critical questions.In recent years the SFRT research community has focused their efforts on better understanding the correlations between the physical parameters and the biological response in SFRT.
One of the first attempts in that direction was the work of Regnard et al (2008).They assessed the impact of different ctc distances on the balance between sparing and curing and concluded that larger ctc (200 μm instead of 100 μm) provided a larger widening of the therapeutic window.In a continuation study, Serduc et al (2009) attempted to assess the influence of the microbeam width at constant valley dose.However, since the same ctc was used in all the several configurations, the valley width was very different in all groups, and, thus, no clean conclusions can be extracted from this study since several parameters were varied at the same time.
In a carefully designed study by Rivera et al (2020), Fischer 344 rats with fibrosarcoma tumor allografts were irradiated with kV x-rays using a small animal irradiator with different configuration SFRT collimators to assess the impact on tumor control of a large range of radiation spatial fractionation parameters (e.g.peak, valley doses, PVDR, etc).The dosimetry parameters most closely associated with tumor response were tumor EUD, valley dose and percentage tumor directly irradiated.Average dose and peak dose showed the weakest associations to tumor response.Only the uniform radiation group did not gain weight post-radiation, indicative of treatment toxicity; however, body weight change in general shows weak association with all dosimetry parameters except for valley (minimum) dose, valley width, and peak width.The finding that peak dose lacks correlation with treatment response in the rat model study directly challenges the clinical practice of using peak dose to prescribe SFRT treatment.The question is whether the finding is specific to one study or can be generalized to singlefraction preclinical SFRT studies?To answer this important question Fernadez, Chang, and Prezado reviewed all SFRT preclinical studies available up to 2022 including some unpublished data (Fernandez-Palomo et al 2022).Of the 16 preclinical studies that met the review criteria (single SFRT treatment, adequate dosimetry data, and having a control group) there is a large variety of SFRT types (microbeams, minibeams, x-rays and protons) and tumor types (different brain tumors and fibrosarcoma).Increased life span (ILS) from the SFRT treatment compared to the life span of the untreated control group is used to evaluate treatment response.No strong correlation was found between ILS and any of the dosimetry parameters when all data were included in the analysis, regardless of the preclinical SFRT mode used.However, strong correlations appeared when MRT and MBRT were analyzed separately, which might suggest potentially different modes of action in both techniques.In both cases, and despite the heterogeneity of the data, valley doses stand out as one of the key parameters in terms of ILS, whereas peak dose was only weakly correlated.Nevertheless, we need to exercise caution in translating preclinical findings to clinical applications.The peak dose may be an important clinical and preclinical SFRT dosimetry parameter for cytotoxic cell killing and for triggering important secondary radiobiological processes, such as vascular permeability or immune cell infiltration, all of which merit further investigation.There are also important differences between preclinical SFRT studies and clinical SFRT applications.The size of the peak and valley width (in cm) is significantly larger in clinical SFRT than in preclinical studies (10-100 s microns); the PVDR is significantly lower (~5) in clinical SFRT than in preclinical studies (>10 in MBRT and 20-50 in MRT).More importantly, clinical SFRT is often followed by a course of conventional RT, whereas animal studies often have a single SFRT regime.Additionally, the fact that cancer patients may respond very differently to the same treatment compared to animal models is also a well-known limitation of preclinical studies.

Discussion and main knowledge gaps
In the previous sections of this review article, we have illustrated that vast amount of knowledge on SFRT has been acquired through decades of clinical application and preclinical research.Our understanding of SFRT has exponentially increased in recent years, after SFRT recaptured the attention of the field.Many published and unpublished clinical data on GRID therapy and LRT showed promising tumor and symptom control.In clinical application, GRID and LRT are generally delivered in combination with a course of conventional RT to achieve higher response rates.As the conventional radiation delivers uniform dose, it can be viewed as a boost to valley dose.This is consistent with the finding of valley dose has strong correlation with tumor control (see section 4.2).
On the other hand, numerous preclinical studies, most of which use SFRT monotherapy at small scales (MRT and MBRT), have also revealed a rich collection of mechanistic information on SFRT (see section 3), and SFRT's capacity as a safe and effective priming therapy not only for radiotherapy but for chemotherapy and immunotherapy.More limited but compelling data have also exhibited that as a monotherapy SFRT can safely and eradicate aggressive tumor models (Prezado et al 2018, Prezado et al 2019, Sotiropoulos et al 2021, Bertho et al 2021).The narrow beams in MRT and MBRT and their associated increased normal tissue tolerance seem to enable the safe use of high enough valley doses without the need of combining it with conventional RT.
SFRT appears promising for improving local control of radioresistant tumors (e.g.glioblastomas) (Tsien et al 2009) and of entities in which local control is a surrogate for survival such as Ewing sarcomas (Rodriguez-Galindo et al 2002).Brain tumors and especially glioblastomas can be seen as a good tumor candidate for future trials with SFRT.Indeed, high-grade gliomas are associated with poor local control and a high risk of toxicity (e.g.radiation necrosis and cognitive impairment) when treated with conventional RT.On the other hand, SFRT could be a game changer in diseases generally cured by RT but with a high rate of long-term toxicities (e.g.some pediatric tumors, reirradiation cases).When considering its biological effect, SFRT could be exploited for precision medicine.Some authors consider that some tumor regions may be more radioresistant than others and exhibit features detectable in functional imaging.These tumor regions could be targeted by SFRT while the rest of tumor is treated (or not) by conventional RT.This strategy was described by Ferini et al in a study where 30 patients with bulky lesions received a LRT session (median dose, 15 Gy) at the interface between tumor regions of 18F-FDG (fluorodeoxyglucose F 18) high and low uptake and a hypofractionated RT course treating the whole target 7 d after LRT (Ferini et al 2022).The rate of symptomatic response was 100% within the first weeks; complete tumor response was described in 5/30 patients with a median follow-up of 10.75 months.While all underlying biological mechanisms are not fully understood, the triggering of an anti-tumor immune response by SFRT opens the way to combined treatment with immunotherapy including checkpoint inhibitors, especially in tumors known to be sensitive to immunotherapy such as melanoma and non-small cell lung cancers (Bertho et al 2023).
Despite the considerable knowledge gained so far, we still have a long way to go to reach the understanding necessary to fully unleash the power of spatially fractionated radiation therapy for patient care.This would require bridging the gap in two main aspects: (1) A deeper knowledge of the SFRT radiobiology (see section 3), both at molecular and tissular level.Some main open questions are reflected in table 2 and include unraveling the distinct pathways of immune activation after SFRT, how the SFRT vascular impact correlates with immune response, how bystander effects are activated in SFRT, how they correlate with the rest of the effects and especially a deeper insight in terms of the biologic mechanisms involved in normal tissue sparing.The comprehension of which biological mechanism(s) dominate or their relative weight for a given SFRT treatment application is also lacking.
(2) A deep understanding of correlation between the many dosimetric and geometric parameters of SFRT and the subsequent treatment response, which can be distinctly different than those in conventional radiation therapy.This is a challenging goal due to the high complexity of SFRT dosimetry, as explained in section 4.2., which is further complicated by the fact that most clinical SFRT treatments are followed by a conventional course of radiation therapy.Knowing the correlation between dosimetry parameters and treatment response is essential.We can and should advance our understanding in parallel on both the above correlation and the underlying mechanisms of SFRT.In this path the fact that each technique (GRID, LRT, MBRT and MRT) have their own working range of parameters (see table 1) in which the best compromise between beam widths and therapeutic doses can be found should be taken into consideration when designing both clinical trials and experiments.
To disentangle the above correlation two different strategies could be used: (I) To conduct extensive clinical trials that measure treatment outcome for a given SFRT configuration in term of dose and geometry.This will be extremely time consuming since standardized treatment parameters are used in most clinical trials to minimize dosimetry variation in patients.In the SFRT case, this would lead to a close self-correlation among all the dosimetry parameters in SFRT treatment.For instance, the correlation between peak width and valley width in such a clinical trial could be 1.0, and for a given prescription dose (often at peak) the peak dose and valley dose can also have a non-negligible correlation.Therefore, although clinical trials of such design may reveal the correlation between SFRT treatment outcome and dosimetry as a whole, it would not be able to identify which SFRT dosimetry parameter(s) have more clinical significance than others.
(II) Free of the constraints in clinical trials, preclinical studies are ideal for investigating correlations between SFRT dosimetry parameters and treatment response and shed light on clinical application.Section 2.2.has illustrated many benefits of SFRT preclinical studies.The last important gap in SFRT research perhaps lies in the clinical relevance of preclinical research.There are important differences between preclinical SFRT studies and clinical SFRT applications.The geometrical scale is significantly larger in clinical SFRT than in preclinical studies while the spatial modulation is lower.More importantly, clinical SFRT is often followed by a course of conventional RT, whereas animal studies often have a single SFRT regime.If SFRT as monotherapy is not successful at local control it does not mean it is not a successful priming therapy for the following treatment of radiotherapy, chemotherapy, or immunotherapy.More preclinical studies conducted under SFRT clinical application conditions are needed to narrow the knowledge gap between bench to bedside.
Once we have bridged those gaps, we will be able to provide a more predictive dose prescription method and to integrate relevant radiobiological modeling in the treatment planning systems.The latter could be based on the most relevant dosimetry parameter or in a pareto front after a multi objective optimization.In addition, it is of paramount importance to keep advancing strategies for dose calculation in MRT and MBRT clinical applications.These strategies should employ the development of specific modules that include appropriate physics and scoring parameters for each application.In addition, they should allow the inclusion of micro and minibeam production methods (i.e.collimators) or, alternatively, accurate beam models resembling the collimation method.Finally adapted TPS modules should include optimization objectives that describe MRT and MBRT dose distributions.
Finally, in contrast with the once new radiation therapy treatments such as IMRT, VMAT, and proton therapy, which are all based on the same radiobiology that most radiation oncology professions are trained on, SFRT directly challenges it, and it might explain some hesitation in the community.SFRT education to the radiation oncology community is crucial for bringing the potential benefits of SFRT to more cancer patients.

Conclusions
SFRT is a paradigm shift from uniform dose conventional RT.Several decades of clinical use and preclinical studies have generated mounting evidence that suggest SFRT has the potential to significantly increase radiotherapy therapeutic index, especially for clinically challenging scenarios.The ongoing advancement of both SFRT technology and radiobiology knowledge is generating a growing interest in SFRT worldwide, newly formed working groups on guidelines and protocols and more clinical SFRT trials are ongoing or being planned.The results obtained thus far both in preclinical experiments and clinical treatments suggest that SFRT has several potential advantages.First, it can potentially increase the radiation dose to radioresistant regions within the tumor, improving local tumor control.Second, it may spare nearby critical structures, such as organs at risk or normal tissues that are sensitive to radiation.Third, SFRT can overcome the limitations of tumor heterogeneity, where different regions of a tumor may respond differently to radiation.By selectively targeting specific regions, SFRT may potentially improve treatment outcomes.
However, SFRT today has not yet reached its full potential.This would require a deeper and global comprehension of the unique radiobiology in SFRT, and especially, how the biology correlates with the multidimensional and multiscale nature of SFRT.Further biological experiments are urgently needed to precisely parametrize the relationship between the irradiation parameters (beam width, spacing, PVDR, peak and valley doses) and radiobiology.
Along this line, SFRT is an ideal approach to illuminate the dark landscape of this vast terra incognita of how the physical parameters of the irradiation modify the biological response, and thus, SFRT may offer enormous opportunities for the creation of optimal patient treatments.Prudence needs to be exercised when translating the preclinical results to patient treatments though.
The authors of this review encourage all interested scientists to join us in the exploration of SFRT.The variety of SFRT clinical and preclinical technical implementations described in section 4.1, most of which are achievable at reduced costs at typical radiotherapy clinics and radiobiology laboratories, allow that any clinician or researcher interested could easily join the SFRT community efforts.We believe that we are currently experiencing a very exciting period for SFRT as it is one of a promising and intriguing approach with potential to reshape RT.
. Currently, it is not totally clear how those geometric and dosimetric parameters correlate with both tumor control and normal tissue sparing in SFRT (Fernandez-Palomo et al 2022).Thus, SFRT is a perfect example of the vast terra incognita on how the physical parameters of irradiation can influence the biological response (Prezado 2022).

Figure 2 .
Figure 2. Upper row left: photographs of a commercial GRID collimator (Zhang et al 2022).The image is provided by the vendor.Upper row right: 3D visualization of an LRT plan.Taken with permission from (Blanco Suarez et al 2015).Lower row left: Photograph of a proton minibeam collimator at the Orsay Proton therapy center.Taken from (Flynn et al 2023).Lower row right: rendering of a flexible x-ray minibeam collimator (Stengl et al 2023).The 3D printed scaffold (1) holds the metal and plastic plates (2) screwed by three locking screws (3).Taken from (Stengl et al 2023).
(Fernandez-Palomo et al 2020).Equivalent or superior tumor control compared to conventional RT has been observed in MBRT (Prezado et al 2019, Lamirault et al 2020a, Bertho et al 2021), showing a remarkable widening of the therapeutic window.Proton MBRT has been shown to maintain or enhance tumor control in glioblastoma rat models while lowering the neurological toxicities associated to high doses of radiation (Prezado et al 2019, Lamirault et al 2020b, Bertho et al 2021, Prezado et al 2017b).Rats treated with pMBRT showed significantly less indications of neurotoxicity at the same curative dose of conventional proton therapy (e.g.lower brain tissue necrosis, astrogliosis and shorttime memory impairment (Lamirault et al 2020b)).Additionally, remarkable proportions (up to 83%) of tumor eradication in glioma bearing rats have been observed (Bertho et al 2021) despite using heterogeneous dose distributions contradicting the paradigm of conventional RT.
al 2010, Sabatasso et al 2011, Bouchet et al 2015), cell signaling effects (Asur et al 2012, Asur et al 2015), inflammation and immunomodulatory effects (Bouchet et al 2013, Potez et al 2019, Bazyar et al 2021, Bertho et al 2022b) including abscopal effects at the tumor and healthy tissue level, stem cell migration (Dilmanian et al 2002), free radical production and diffusion covering the valley regions in the tumors (Dal Bello et al 2020).We will briefly describe several proposed mechanisms and the existing experimental evidence.
al 2018, Day et al 2021), variance-reduction techniques (Martinez-Rovira et al 2012), parallelization methods or graphical processing units.In addition, MC codes typically used in MRT and MBRT research do not allow treatment plan optimization which considers the characteristics of MRT and MBRT dose distributions.Current published practices for the calculation of dose distributions in examples of clinical patients are based on, first, optimizing dose distributions in seamless conditions with current clinical TPS and, then, calculating the dose distributions of the optimized treatment plan using MC simulations which include collimator devices at the end of the beam path and appropriate simulation parameters (Ortiz 2022).
(Martínez-Rovira et al 2012), modifying the distributions of dose deposition.Simplifications in the source modeling may lead to inaccuracies in the assessment of MRT (Nettelbeck et al 2009) and MBRT (De Marzi et al 2018) dosimetry parameters.Therefore, the characterization of the source and the development of an accurate beam model are essential components of dose computation engines (Martínez-Rovira et al 2012, De Marzi et al 2018, Dipuglia et al 2019).Whereas reproducing these

Figure 3 .
Figure 3. Impact of a tile angle between the collimator and beamline of 0.25 degrees in pMBRT lateral dose profiles.
al 2017, Prezado et al 2017a), medical linear accelerators (Kundapur et al 2022), proton therapy centers (Peucelle et al 2015), carbon (Martínez-Rovira et al 2017) and Ne ion (Prezado et al 2021) centers.x-ray MBRT requires the use of suitable collimators.In contrast, charged particle MBRT could benefit from magnetic focusing (Girst et al 2016, Schneider et al 2020) by means of new nozzle designs, more compact and with a shorter focal length

Table 2 .
(Grams et al 2022) Grams and Zhang 20232017)FRT and conventional RT and main open questions.What is the role of ROS in the biological mechanisms of SFRT?Is there a link with bystander-like effects?SRS and SBRT treatments, and accordingly, the same dosimeters could be used(Grams andZhang 2023, Grams et al 2023).Treatment planning systems (TPS) are generally quite capable of accurate dose calculations at this length scale, and therefore a TPS which has been commissioned for use in stereotactic body radiation therapy (SBRT) or stereotactic radiosurgery (SRS) treatments should also be suitable for either GRID or LRT.Published guidelines for SBRT/SRS(Smilowitz et al 2015, Halvorsen et al 2017)treatment planning and delivery should also be followed for GRID and LRT dosimetry.Concerning GRID, the collimator should be incorporated within the user's TPS(Grams et al 2022, Grams and Zhang 2023, Grams et al 2023).If this is not feasible, reference lookup tables can be used for estimation of relevant dosimetric parameters(Grams et al 2022).Details on the commissioning and dosimetric characterization of physical blocks for SFRT treatment can be found in the literature(Nobah etal 2015, Zhang et al 2020).Several different LRT planning methods have been used clinically (Wu et al 2020, Grams et al 2021, Duriseti et al 2022).