Shoot-through proton FLASH irradiation lowers linear energy transfer in organs at risk for neurological tumors and is robust against density variations

Objective. The goal of the study was to test the hypothesis that shoot-through FLASH proton beams would lead to lower dose-averaged LET (LETD) values in critical organs, while providing at least equal normal tissue sparing as clinical proton therapy plans. Approach. For five neurological tumor patients, pencil beam scanning (PBS) shoot-through plans were made, using the maximum energy of 227 MeV and assuming a hypothetical FLASH protective factor (FPF) of 1.5. The effect of different FPF ranging from 1.2 to 1.8 on the clinical goals were also considered. LETD was calculated for the clinical plan and the shoot-through plan, applying a 2 Gy total dose threshold (RayStation 8 A/9B and 9A-IonRPG). Robust evaluation was performed considering density uncertainty (±3% throughout entire volume). Main results. Clinical plans showed large LETD variations compared to shoot-through plans and the maximum LETD in OAR is 1.2–8 times lower for the latter. Although less conformal, shoot-through plans met the same clinical goals as the clinical plans, for FLASH protection factors above 1.4. The FLASH shoot-through plans were more robust to density uncertainties with a maximum OAR D2% increase of 0.6 Gy versus 5.7 Gy in the clinical plans. Significance. Shoot-through proton FLASH beams avoid uncertainties in LETD distributions and proton range, provide adequate target coverage, meet planning constraints and are robust to density variations.


Introduction
An increasing number of patients worldwide are being treated by proton therapy due to its increased availability and its superior dose shaping compared to photon therapy (Baumann et al 2020, Chen et al 2023).Contrary to photon therapy, proton therapy is accompanied by uncertainties in linear energy transfer (LET) and relative biological effectiveness (RBE) distribution (Grassberger et al 2011, Paganetti et al 2019) as well as range uncertainties (Gomà et al 2018, Taasti et al 2018a).The range uncertainty requires a larger area to be treated to ensure coverage, exposing more healthy tissue to high doses.LET and presumably RBE are the highest at the end-of-range (Grassberger et al 2011).Therefore, healthy tissue located distally from the tumor, could be inadvertently endangered.Sørensen et al observed an increase in early skin damage in mice of which the skin was irradiated at the distal edge of the Bragg peak compared to the center of the spread-out Bragg peak, while the physical dose was lower at the distal edge (Sørensen et al 2017).High LET D values at the distal edge of the Bragg peak could be a potential explanation for the occurrence of unexpected rib fractures in breast patients (Wang et al 2020), lung density changes in chest wall patients (Underwood et al 2018) and late radiation-induced MRI contrast-enhancing brain lesions in patients with neurological tumors, when treated with proton therapy (Peeler et al 2016, Bahn et al 2020, Harrabi et al 2022).While these contrast enhancing brain lesions generally do not lead to symptoms, their presence reflects early tissue damage and could hamper interpreting image changes in follow-up imaging.
With the increasing use of dual energy CT (DECT) (Almeida et al 2018, Taasti et al 2018b) and the advent of proton-based CT (Dedes et al 2019) and LET measurements (e.g.Stasica et al 2023), the uncertainty on both range and RBE is becoming smaller.In this work, we explore another way to curb these uncertainties.Contrary to classic proton therapy, shoot-through beams position the Bragg peaks behind the patient.A single shootthrough beam deposits dose along its entire track with a slight gradient in dose deposition (figure 1).While a shoot-through planning strategy does not leverage the favorable dose deposition characteristics of protons in the distal region of the target, it could have alternative benefits.By delivering the dose through the plateau region before the Bragg peak, uncertainties on LET and proton range are bypassed (Verhaegen et al 2021).The high energy required inherently leads to sharper lateral beam penumbras and less angular spread.Finally, the proton beam exiting the patient could be used for in vivo dosimetry (Testa et al 2013, Johnson 2018, Verhaegen et al 2021).
Regardless, the lack of distal dose fall-off behind the target in shoot-through beam plans generally would not be favored over the traditional deposition of Bragg peaks inside the tumor.However, with the advent of FLASH radiotherapy, the less favorable dose distribution could be counteracted at least partially by the protecting effect of dose rates exceeding >40 Gy s −1 (ultra-high dose rate (UHDR) in the remainder of this article), illustrated in figure 1  Depending on the FLASH protective factor (FPF), shoot-through proton treatment could be a viable alternative for classic proton therapy with some significant technological and therapeutic benefits.E.g.UHDR are more easily achievable for shoot-through beams, as time-consuming energy switching is not needed (Zou et al 2021) and the omission of an energy-modifying device allows for a higher beam output, a less spread-out beam and a more efficient beam delivery.Moreover, as the dose sparing of organs at risk is not dependent on the proton range for shoot-through beams, plans would be very robust to density changes.This is beneficial for patients with head and neck tumors or certain neurological tumors, as for example changes in nasal cavity filling, could be a reason for plan adaptations.
While the number of technical solutions to reach UHDR keeps growing (Schulte et al 2022), we have decided to focus on the more common pencil beam scanning (PBS) proton accelerators.Alternative methods to make UHDR more readily achievable while maintaining the Bragg peak, are the use of ridge filters in passive scattering systems (Durante et al 2018, Buonanno et al 2019, Diffenderfer et al 2020) or a combination of a universal range shifter and range compensator for pencil beam scanning systems (Kang et al 2022).Zhang et al compared the use of passive scattering with ridge filters to PBS, either in shoot-through or Bragg peak deposition mode, in terms of dose and delivery time (Zhang et al 2021).As these methods would not lead to lower LET-values, they are not directly considered in this study.
In the study presented, the goal was to investigate the feasibility of exploiting the FLASH protective effect to generate clinically acceptable plans using shoot-through proton beams with a lower RBE and lower range uncertainty than standard, clinical plans.Dose averaged LET (LET D ) is evaluated as a readily computable surrogate for RBE (  (McMahon et al 2018).Therefore, a planning study was performed for 5 neurological tumor cases in which LET D was computed and the robustness for density uncertainties was evaluated.Several FPFs were applied for normal tissues outside the CTV to explore which factor would still yield acceptable results.

Patient selection
Five neurological cancer patients treated in our centre between 2019 and 2021 were retrospectively selected based on dose to the hippocampi as organ at risk (OAR) (4 patients, mean dose > 12 Gy) or nearby critical organs (1 patient, optical nerve and chiasm) in the clinical proton plan.The group consisted of 4 glioma patients and one hemangiopericytoma patient.Fractionation schedules ranged from 28-33 × 1.8 Gy as per national and international guidelines (Weller et al 2021).

Treatment planning
For each patient, a previously prepared clinical spot scanning intensity modulated proton therapy (IMPT) plan was available, made by the radiation therapist at the time of patient treatment and evaluated according to the clinical goals listed in table 1.The proton energies used in the clinical plans ranged from 33 to 156.1 MeV.Additionally, a new shoot-through plan was created.For both plans, the treatment planning system RayStation was used with a Mevion Hyperscan S250i beam model with dynamic beam collimation (Mevion Medical Systems, USA).LET D calculations were performed in RayStation version 9A-IonPG and plan optimization and dose calculations were done in RayStation 8A or 9B (RaySearch Laboratories AB, Sweden).The RayStation Monte Carlo dose calculation algorithm was used and an RBE of 1.1 was used in the whole patient.It transports primary protons and secondary protons, deuterons and alpha particles.Heavier fragments and neutrons are neglected.The dose calculation algorithm did not change between versions 8A and 9B.
For the shoot-through plan, only 227 MeV PBS beams were used, as this is the highest proton energy the Mevion Hyperscan S250i produces.A lower energy would suffice for shoot-through of a patient's head, however, it would result in the insertion of one or more Lexan range modulation plates in the beam line, lowering the maximum dose rate.Additionally, higher energies result in a flatter dose distribution which is beneficial for meeting OAR constraints behind the tumor.For each patient, the fractionation schedule matched that of the clinical plan.To improve the technical feasibility of UHDR, plans were made with a maximum of 5 non-parallel opposed beams.For the same reason, a minimum interspot distance of 4 mm and minimum spot weights of 0.5 MU/fraction were used.Using only a single energy layer and a predefined spot grid, plans were optimized using clinical objectives through the standard optimization utility of the treatment planning system.More details on the plan optimization can be found in the Supplementary Information.
A FPF of 1.5 was applied for normal tissues outside the CTV, using MATLAB R2018B.This choice was based on the observation that most FPFs found in literature are within the 1.4-1.8range when a protective effect is observed and quantified (Wilson et al 2020).In case of overlap between the CTV and the OAR, the FPF was not applied.Plans were evaluated according to the clinical goals listed in table 1.To gain insight in the importance of the choice of the FPF, as values reported in literature can vary, clinical goals in the nominal plans were evaluated for additional FPFs ranging from 1.2 to 1.8.

Linear energy transfer
LET D distributions were calculated for the clinically administered proton plan and the shoot-through plan using RayStation 9AR IonPG, previously validated (Wagenaar et al 2020).RayStation 9AR IonPG computes doseaveraged LET in water, normalized to unit density according to equation (1): with S E el i ( ) the unrestricted electronic stopping power of protons with kinetic energy E, D E, z i ( )is the absorbed dose-to-water contributed by protons with kinetic energy E at location z of the ion i.The LET D is calculated for all primary and secondary protons.A dose threshold of 2 Gy for the total treatment was applied, as the clinical relevance of LET D for lower doses is considered minor.The LET D distributions were evaluated on their average LET D and maximum LET D in 2% of the volume (LET D2% ).

Robust evaluation
The effect of CT density uncertainties on the dose distribution were investigated by robust evaluation in RayStation 8B or 9B.The dose was recalculated for two scenarios: one with an overall 3% lower density and one with 3% higher density.Subsequently the voxel-wise minimum and maximum dose were calculated, implying Table 1.Clinical goals stated as equivalent dose in 2 Gy per fraction (EQD2).OAR = Organ at risk.For the shoot-through plans, several FLASH protective factors (FPF) outside the CTV were used.For both the shoot-through and clinical plans, the number of plans for which the clinical goals were met, are stated both for the nominal dose distribution as well as the voxelwise maximum or minimum dose distribution for the clinical and the shoot-through plan with FPF 1.5.In case a bilateral organ is listed, the clinical goal is considered to be met in case it is met for both organs separately.Clinical goals listed in italics are soft constraints.Bold numbers indicate a difference in the number of clinical goals met depending on the FPF.As the selected patients were difficult cases, the coverage of the target was compromised in 2 cases to spare nearby critical structures.that for every voxel, the minimum and maximum dose out of the three scenarios (nominal, 3% higher and 3% lower density) was taken.The robustness of the target coverage was judged by the V95% and D98% of the CTV in the voxel-wise minimum dose for the evaluated scenarios.The robustness of the OAR sparing was evaluated on the D2% of the voxel-wise maximum dose for the evaluated scenarios.Additionally, the robustness evaluation was repeated for 14 scenarios comprising of 1 mm shifts (6 principal directions and 8 vertices) and adding positive and negative density changes of 3%, resulting in 28 scenarios in total (2 × 14), according to the consensus within the Dutch Proton Therapy (DUPROTON) group (Korevaar et al 2019).The 1 mm represents the residual setup and delineation uncertainty with online imaging and the 3% density changes are introduced to take into account range uncertainties for single energy CT scans (Mcgowan et al 2013).

Dose distribution
Examples of the dose distribution for both the clinical and the shoot-through plan with a FPF of 1.5 are depicted in figure 2. Target coverage was comparable for both strategies (V 95% differences 1.7%).Although less conformal, shoot-through plans met the same number of clinical goals as the clinical plans (in the nominal dose distribution, table 1).Dose volume histogram (DVH) parameters are shown in table 2. Doses to the brainstem, hippocampi excluding the CTV (hippocampi-CTV), pituitary and optical nerves were lower in the shootthrough than in the clinical plans.The dose to the supratentorial brain was higher for the shoot-through plans in 3/5 patients.Notably, for all patients, at least a few organs of the auditory and visual system received higher doses for shoot-through plans, yet still met the same clinical goals as the clinical plans except for one occasion: in patient 4, the optical chiasm D 0.03cc and the D 40% for both hippocampi were only met in the clinical plan.In other patients, the clinical plan did not meet one or more clinical objectives (all soft constraints) which were met by the shoot-through plan.Additional FPFs ranging from 1.2 to 1.8 were considered.For most OAR, while a difference in dose was observed, the same number of clinical goals were met (table 1).The reason for this is that most clinical goals are considering near-maximum doses D 0.03cc and these high-dose areas are generally located in the part of the OAR overlapping with the CTV, where the FPF was not applied.
The shoot-through plans with FPF 1.5 were in some cases less robust to the combination of 1 mm displacement and 3% density perturbation than the clinical plans (table 1).

LET D
For the clinical plans, large variations in LET D are observed surrounding the CTV, while LET D for shootthrough plans mostly showed variations in bone, likely related to an increased yield and different energy of secondary protons (e.g.bone versus soft tissue) (Grassberger et al 2011).Examples of LET D distributions for both the clinical and the shoot-through plan are depicted alongside the dose distribution in figure 2.
For shoot-through plans, the LET D maximum 2%, is a factor 1.2-8 lower than for clinical plans for all critical structures.The average LET D is lower and less variable for the shoot-through plans (figure 3), e.g. for the brainstem, an organ nearby all tumors that were considered, the average ± standard deviation LET D is 0.9 ± 0.2 keV μm −1 for the shoot-through plan and 4.0 ± 0.5 keV μm −1 for the clinical plans.For the optical chiasm, the average LET D is 0.94 ± 0.03 keV μm −1 and 3.6 ± 0.8 keV μm −1 for the shoot-through and clinical plans, respectively.

Robustness to density variations
Figure 4 illustrates the effect of density uncertainties on the dose distribution.The differences in D2% dose to the OAR never exceed 0.57 Gy for the FLASH shoot-through plans in the voxelwise maximum dose distribution and the maximum drop in CTV D98% is 0.20 Gy.For the clinical plans, the maximum drop in CTV D98% is 0.56 Gy and the D2% doses in the voxelwise maximum dose distribution show more variable differences with the highest differences seen in the hippocampus (maximum difference 3.04 Gy) and cochlea (5.72 Gy), although this is highly dependent on the anatomy and beam arrangement.

Discussion
For protons, an RBE of 1.1 is generally assumed for all of the deposited physical dose, along the entire proton track.However, it is also broadly accepted that this is a simplification as the RBE is a function of dose, LET, cell type (α/β) and the considered endpoint (Paganetti et al 2019).While the other factors play a non-negligible role, LET serves as a main input parameter for a number of RBE models in addition to the dose (Grassberger et al 2011, Paganetti et al 2019).A LET D -weighted dose displays a roughly linear relationship with RBE   However, in this respect, there are two shortcomings to our study.First of all, for the CTV, the average LET D values for the shoot-through plans are 0.95 ± 0.02 keV μm −1 , while 2.5 ± 0.1 keV μm −1 for the clinical plans.This raises the question whether the RBE of 1.1 used in this study for the shoot-through plans is justifiable.Although generating clinically acceptable plans could become more challenging when assuming a lower RBE, the conclusions on LET D and robustness would still hold.Secondly, we recognize that the relationship between LET D and RBE requires careful consideration if the dose levels of the standard and FLASH shoot-through plans would become very different, e.g. in case hypofractionation would be considered for FLASH shoot-through plans.Regardless, considering the range uncertainty for the clinical plans in combination with the steep LET D gradients near OARs, a shoot-through approach would lead to a more robust plan in terms of LET D deposition.
For clinical proton beams, range uncertainties of 3%-5% are generally assumed, partially caused by anatomical changes and CT conversion uncertainties (Paganetti 2012).The dose distribution for FLASH  shoot-through beams of 227 MeV is less dependent on the tissues they traverse, as the gradient of the dose deposition is around 1.7 %/mm (Verhaegen et al 2021) and the position of the Bragg peak is irrelevant as long as it is outside of the patient.For patients with neurological tumors, this is especially relevant as patients with tumors close to sinuses and cavities could be deemed not eligible for proton treatment.This could be due to differences in sinus filling, which could be detrimental for the dose distribution, or because they need frequent and time-consuming imaging and adaptation for this reason.Shoot-through FLASH plans circumvent this risk, as was reflected by the robustness of shoot-through plans to 3% density changes.While this obviously deteriorates the conformity of the dose distribution to the tumor, we have shown that it was still possible to produce clinically acceptable plans, considering a FLASH protective factor of 1.5 or even as low as 1.2 for most OARs considered.
FLASH shoot-through plans suffer from a number of planning difficulties.The use of shoot-through beams for FLASH radiotherapy with protons limits the number of optimal beam angles.As clinical plans deposit the Bragg peak inside the tumor, beams can be tilted more cranio-caudally without leading to dose deposition in e.g. the oral cavity.For shoot-through plans more cranially tilted beams could still be part of an acceptable treatment plan, but then dose constraints from head and neck plans should be taken into consideration.In this study, beams were limited to a maximum angle of ~20 degrees with the axial plane.Using a single energy layer limits the degrees of freedom and OARs are more difficult to evade due to the shoot-through nature of the beams.Beam collimation with a dynamic aperture poses a challenge to achieve high dose rates, but the dynamic aperture could be replaced with static blocks to sharpen the lateral penumbra of the pencil-beam scanned fields.Moreover, the lateral penumbra is inherently smaller for high energy shoot-through beams as there is less lateral scattering.While this planning study is too small to give a rigorous dosimetric comparison between shoot-through plans aided by a FLASH protection effect versus standard proton plans, we believe this planning study showed that even without beam collimation, acceptable plans could be within reach.Wei et al have demonstrated that acceptable lung stereotactic irradiation plans can be made with 3 or 5 shoot-through beams, even without taking the FLASH protective factor into account (Wei et al 2022).
The most important reasons to even consider introducing FLASH radiotherapy in future daily practice are twofold.The first reason would be an increase in local control through FLASH irradiation, which could be obtained by increasing the total dose to the tumor without increasing the normal tissue complications, or by using altered fractionation schedules like hypofractionation or increasing the target volume.There is a hesitance in the application of hypofractionation or higher total dose (Wegner et al 2019) due to the possible unacceptable and irreversible increase of central nervous system side effects like radionecrosis.The already large CTV margin of 1.5-2 cm (Niyazi et al 2016) used for e.g. a glioblastoma includes many still functioning healthy cells limiting the total dose and volume, although the margin is often not enough due to the wide spread growth of these tumors.If these functioning cells in the margin region would be able to survive due to the FLASH effect a possible larger area could be targeted with a higher dose.
The second reason to consider the use of FLASH radiotherapy in daily practice would be the decrease of side effects.Since one of the most feared side effect is cognitive decline, sparing cognition related OARs (Eekers et al 2018) could be a relevant goal, possibly increasing quality of life.
It can be debated whether the currently used fractionation schedule (28-33 × 1.8 Gy, as per national and international guidelines) would result in a FLASH effect, due to the reported but uncertain dose threshold per fraction to elicit the FLASH effect.Montay-Gruel et al have shown a decrease in memory loss due to FLASH irradiation for a 3 × 10 Gy schedule while it was not observed for a 4 × 3.5 Gy schedule (Montay-Gruel et al 2021).In the Netherlands, a model-based approach is used to select patients with neurological tumors that benefit most from proton therapy, based on the dose to the hippocampi and the supratentorial brain (Langendijk et al 2013, van der Weide et al 2021).This favors tumors with large dimensions, reflected in the patient cohort used in this study (largest dimension ranging from 5 to 14 cm), which limits the options for hypofractionation and would be challenging to irradiate with UHDR considering the current state of technology (Zou et al 2021, Krieger et al 2022).We expect that smaller tumors treated with a hypofractionated scheme would be more eligible for FLASH treatment.Regardless, the current LET D results and the robustness to range uncertainties would still translate to hypofractionated treatments.

Conclusions
Proton shoot-through FLASH beams treating neurological tumors avoid LET D and range uncertainties, while still providing adequate target coverage and meeting planning constraints if the assumed FLASH protective factor is in the order of 1.5 or even as low as 1.2 in most cases.Future research should focus on the feasibility of creating clinically acceptable hypofractionated shoot-through plans and take into account the newest insights on FLASH protective factors, required technical conditions and the LET D -RBE relationship at UHDR.
Paganetti et al 2019).McMahon et al showed that LET D weighted dose correlated with RBE

Figure 1 .
Figure 1.Effective dose profile for shoot-through beams for regular and ultra-high dose rate (UHDR).In the latter case, a FLASH protective effect in both proximal and distal regions is assumed.Adapted from figure 1, (Adapted from Verhaegen et al 2021).© 2021 Institute of Physics and Engineering in Medicine.All rights reserved.

Figure 2 .
Figure2.Examples of dose distributions and LET D distributions for clinical and shoot-through plans of 5 brain tumor patients.For the FLASH shoot-through plans, a FLASH protective factor of 1.5 is applied outside of the CTV (white contour).For the LET D distributions, a 2 Gy total dose threshold is used as LET D in areas with lower doses is not considered meaningful.

(
McMahon et al 2018) although the relationship might be different depending on the exact definition of LET D used (Hahn et al 2022).Since LET D is a physical quantity that can be calculated and corresponds well to the measurable mean lineal energy (Tran et al 2017, Wagenaar et al 2020), we decided to focus on LET D as an indication of the RBE-variations of the plans.Figure 3 showed how limited the LET D -variations (and by proxy, the RBE variations for a certain dose level) are for single energy shoot-through plans.For clinical plans, on the contrary, large LET D values and variations are observed in critical structures in the vicinity of the tumor.

Figure 3 .
Figure 3. Boxplots of (a) volume-averaged LET D for the CTV and OARs and (b) maximum LET D received by 2% of the volume.A total dose threshold of 2 Gy was used for all LET D distributions.In case of bilateral organs, the highest value for both organs is reported.

Figure 4 .
Figure4.Boxplot of the dose difference of either the voxelwise minimum dose (CTV D 98% ) or voxelwise maximum dose (D 2% CTV and OARs) compared to the nominal dose for 3% density changes.In case of bilateral organs, the difference for the organ with largest dose is reported.Shoot-through indicates the dose of the shoot-through plans with a FLASH protective factor of 1.5.

Table 2 .
Dose volume histogram parameters for both the clinical and shoot-through plans.For the shoot-through plans, a FLASH protective factor of 1.5 is applied outside of the CTV.Bold numbers indicate where the shoot-through plan underperforms with respect to the clinical plan.For the rows depicted in italics, the clinical goal is a soft constraint.