Mechanisms and review of clinical evidence for variations in relative biological effectiveness in proton therapy

Abstract

Proton therapy is increasingly being used as a radiation therapy modality. There is uncertainty about the biological effectiveness of protons relative to photon therapies as it depends on several physical and biological parameters. Radiation oncology currently applies a constant and generic value for the relative biological effectiveness (RBE) of 1.1, which was chosen conservatively to ensure tumor coverage. The use of a constant value has been challenged particularly when considering normal tissue constraints. Potential variations in RBE have been assessed in several published reviews but have mostly focused on data from clonogenic cell survival experiments with unclear relevance for clinical proton therapy. The goal of this review is to put in vitro findings in relation to clinical observations. Relevant in vivo pathways determining RBE for tumors and normal tissues are outlined, including not only damage to tumor cells and parenchyma but also vascular damage and immune response. Furthermore, the current clinical evidence of varying RBE is reviewed. The assessment can serve as guidance for treatment planning, personalized dose prescriptions, and outcome analysis.

Introduction

Radiation therapy treatment planning is based on prescription doses and normal tissue constraints that are defined empirically and site specific. When treating patients with protons, the potential difference in biological effect compared to photons for the same dose is considered by applying a relative biological effectiveness (RBE), defined as the ratio of photon versus proton dose to reach the same level of effect. In treatment planning for proton therapy (PT), all treatments are currently prescribed using an RBE of 1.1. This value was agreed upon in the early days of PT as a conservative average for tumor control. The use of a generic, spatially invariant RBE within tumor and normal tissue structures disregards the evidence that the RBE varies with LET (linear energy transfer), physiological and biological factors, and clinical endpoint.

A mathematical expression for the RBE based on the linear-quadratic dose-response relationship is shown in equation (1) [1]. Accordingly, the proton RBE is a function of α_x and β_x of the reference photon radiation, α and β of the proton radiation (depending on the proton LET), and the proton dose per fraction, D_p (neglecting dose rate and time-dependent repair corrections). In this article, LET is reported as dose-averaged LET_d in units of keV/μm [2]. Several empirical RBE models have been developed based on fits to clonogenic cell survival (CCS) data from various cell lines. A comparison of these models showed consistent trends but revealed considerable variations in absolute RBE values [3]. One of these models is shown in equation (2) with fit parameters p₀-p₃ [4].

$$\mathrm{RBE}\left(D_p,{[\alpha /\beta ]}_x,LE{T}_d\right)=\frac{1}{2{\mathrm{D}}_{\mathrm{p}}}\left(\sqrt{{[\alpha /\mathrm{\beta }]}_{\mathrm{x}}^2+4{[\alpha /\beta ]}_{\mathrm{x}}\alpha \left[{\mathrm{LET}}_{\mathrm{d}}\right]/{\alpha }_{\mathrm{x}}{\mathrm{D}}_{\mathrm{p}}+4^{\mathrm{\beta }\left[{\mathrm{LET}}_{\mathrm{d}}\right]}/{\beta }_{\mathrm{x}}{\mathrm{D}}_{\mathrm{p}}^2}-{[\alpha /\mathrm{\beta }]}_{\mathrm{x}}\right)$$ (1)

$$RBE\left(D_p,{[\alpha /\beta ]}_x,LE{T}_d\right)=\frac{1}{2D_p}\left(\sqrt{{[\alpha /\beta ]}_x^2+4D_p{[\alpha /\beta ]}_x\left[p_0+\frac{p_1}{{(\alpha /\beta )}_x}LE{T}_d\right]+4D_p^2{\left[p_2+p_3{\sqrt{\{\alpha /\beta \}}}_{x}LE{T}_d\right]}^2}-{[\alpha /\beta ]}_{\text{x}}\right)$$ (2)

For a 6-MV photon beam, secondary electrons result in an LET of ~0.2 while the LET of protons ranges from ~1–25 in the energy region of interest in PT. With fewer tracks involved for the same dose compared to photons, protons deposit dose more concentrated in space and the clustering of energy deposition mainly in the cellular DNA can result in damage that is more complex and therefore more difficult to repair compared to photon induced damage. The increase in ionization density also causes an increase in reactive oxygen species (ROS) with protons compared to photon radiation [5–7]. The increasing LET with increasing depth of penetration is of particularly concern for critical structures immediately downstream of the target if single-field uniform dose (spread-out Bragg peak, SOBP) is delivered. For multiple fields and particularly in intensity-modulated PT (IMPT) delivering inhomogeneous dose distributions per field, the LET distribution can be highly inhomogeneous. LET values in patients are typically higher than ~10 in the distal fall-off, but only ~2–4 in the center of the target for typical beam arrangements (Figure 1) [8, 9]. Figure 2 shows RBE relationships based on an empirical model deduced from CCS experiments.

Figure 1:

Left side: Typical ranges of linear energy transfer (dashed line; dose-averaged LET_d) in a spread-out Bragg peak (solid line) in water. Right side: Dose (%) and LET_d (keV/μm) in a chordoma patient based on intensity modulated proton therapy. Figure adapted from [9, 21].

Figure 2:

Proton RBE for clonogenic cell survival as predicted by an empirical model (equation 2) [4]. Left: RBE as a function of LET at 2 Gy (solid (α/β)_x=2 Gy; dashed (α/β)_x=10 Gy). The grey area shows the clinically and dosimetrically most relevant region as LET values are typically between 2.5 and 13 keV/μm [21]. Middle: RBE as a function of dose for LET=2.5 keV/μm (solid (α/β)_x=2 Gy; dashed (α/β)_x=10 Gy). The grey area shows the clinically most relevant region for a standard 2 Gy fractionation, i.e., doses to organs at risk <2Gy. Right: RBE as a function of (α/β)_x for a dose of 2 Gy (solid LET=2 keV/μm; dashed LET=10 keV/μm). The grey area shows the clinically most relevant region of (α/β)_x=2–10 Gy.

The RBE for CCS also depends on the cell line. Differences in DNA damage complexity can be associated with differences in repair pathways such as, for instance, a higher portion of homologous recombination versus non-homologous end-joining with higher LET [10–17]. Higher complexity of DNA damage also leads to slower repair [6]. Furthermore, differences in foci size comparing photon and proton irradiation at the same dose indicate significant variations in inflammatory response [18–20]. The analysis of experimental in-vitro data show a trend of increasing RBE as (α/β)_x decreases, i.e., the RBE for late damage is expected to be higher than for early responding tissues (Figure 2) [21], with (α/β)_x indicative of the curvature of the cell survival curve and thus the sensitivity to fractionation.

Furthermore, the RBE for cell survival increases with decreasing dose (Figure 2), which is more pronounced for late responding compared to early responding tissues. A larger dose dependency is predicted for higher values of LET. The dose dependency of the RBE in PT is not well studied experimentally but might be significant for doses below 1 Gy. Note that the linear-quadratic equation might not be valid below ~1 Gy [22, 23] which could affect dose-response estimations for organs at risk typically receiving less than 1 Gy for a 2 Gy per fraction treatment. It may also be questionable (depending on the endpoint) for doses above ~10 Gy [24], which could have implications when considering hypo-fractionated treatment regimens [25–27].

Clonogenic cell survival data and in vivo pathways

It is unclear to what extend these RBE relationships based on CCS should guide clinical practice. Biological mechanisms leading to tumor control or normal tissue toxicities are multifarious and the classical target cell theory [1], which is at the core of the RBE concept, is insufficient to describe effects in patients. Figure 3 schematically and roughly outlines the most relevant aspects. Damage to cells start with either direct or indirect damage that might undergo repair or result in cell inactivation due to apoptotic or necrotic pathways. Due to differences in DNA damage, the ratio of apoptotic versus necrotic pathways likely differs between photon and proton radiation [28]. The diagram is a simplification and disregards many aspects, e.g., cross talks between pathways, programmed necrosis versus secondary necrosis [29], mitotic catastrophe, senescence, autophagy, paraptosis, and other phenomena [30]. For instance, apoptosis and autophagy do contribute to immunogenic response [31] and both necrotic as well as apoptotic pathways may release anti-inflammatory cytokines.

Figure 3:

Schematics of some relevant pathways leading to tumor cell kill and normal tissue toxicities. Starting with direct and indirect DNA damage to the tumor and parenchyma, there are repair pathways that, upon failure, lead to cell kill of tumor cells (left) or damage to functional subunits (right). The text in green indicates pathways related to in vitro cell kill studies. In addition to direct damage to cells, there is radiation-induced damage to vasculature leading to vascular disruption which impacts tumor and normal tissue response. Similarly, adaptive and innate immune response are impacted by radiation. Clinically relevant endpoints are indicated in red. Note that the two empty blue boxes are placeholders for mirrored images of the upper two blue boxes.

With respect to direct (predominantly apoptotic) cell death in tumors we might in part rely on knowledge gained from in vitro data. However, in addition to cell kill, there can be functional impairment in tumor cells, e.g., loss of proliferation while maintaining the ability to release antigens. Furthermore, tumor response does depend on the microenvironment [32]. While patient specificity as well as tumor heterogeneity cannot be fully considered in vitro, they are impacting both photon and proton data.

For normal tissues, there is no direct path to early and particularly late toxicities. Radiation causes the activation of cytokine cascades causing radiation protection or sensitization [33]. The complex pathways will depend on the type of DNA damage and impact response differences between protons and photons. In order to understand normal tissue toxicity we need to understand the inactivation of functional organ subunits [34] as well as the distribution of damage in an organ which, due to the difference in integral dose and dose distribution, differs for the two radiation modalities. Normal tissue sensitivity is determined by tissue organization of these subunits leading to a distinction between parallel and serial organs [35]. This is not covered by RBE but should ideally be considered in NTCP (Normal Tissue Complication Probability) models via organ dose distributions, including potential physiological interactions of different organs (e.g., lung and heart) [36, 37].

Furthermore, radiation damage triggers innate and adaptive immune response. Radiation mediated immune response in normal tissue is part of inflammatory pathways that preludes acute normal tissue toxicities as well as late toxicities via chronic inflammation. Radiation can also introduce immune-suppressive effects via detrimental effects on tumor infiltrating lymphocytes, lymphatic vessels and nodes, as well as circulating lymphocytes [38]. Radiation sensitivity (and likely RBE) differs in lymphocyte sub-populations and even by activation status of lymphocytes [39–41]. The lower integral dose with PT reduces the dose to circulating lymphocytes and might cause differences in lymphopenia between PT and photon therapies [42–48].

Finally, damage to endothelial cells impacts the RBE. Effects on small- and medium- sized vessels leads to disruption of the endothelial lining and the vasculature. Changes in endothelial cell phenotypes cause increased vascular permeability and tumor cell death [49–51]. The role of endothelial cells in tumor response is controversial [49, 50, 52–55]. Endothelial cells are generally assumed to be less sensitive compared to tumor cells [56, 57]. The loss of endothelial cells can lead to a decrease in blood perfusion and an increase in hypoxia triggering increased levels of hypoxia inducible factor-1 (HIF-1) which promotes neovascularization [58, 59]. Vascular abnormalities as presented in tumors negatively impact immune response [60] and damage-associated molecular patterns (DAMPs) activate the innate immune system via endothelial cell activation [61].

In normal tissues, capillaries present the most radiosensitive vessels because the single layer of endothelium is susceptible to ionizing radiation [62]. In addition to cell kill, radiation leads to endothelial cell activation involving expression of cytokines, chemokines and adhesion molecules facilitating recruitment and attachment of circulating leukocytes on the vascular wall [63]. Adhesion molecules are mediators of inflammatory reactions as their expression leads to increased recruitment of immune cells. A higher LET is likely more effective in triggering endothelial cell adhesiveness leading to increased vascular permeability.

In summary, tumor control and normal tissue complications caused by radiation are a combination of four aspects of which only the first two are related to CCS:

1. direct cellular effects: clonogenic cell death of tumor cell populations and normal tissue cell populations that act as functional subunits

2. direct cellular functional effects: inactivation of tumor cell populations or compromising the functionality of functional subunits in normal tissue

3. indirect effects: vascular disruption due to endothelial cell death/damage

4. indirect systemic effects: antigen release, cytokine/chemokine activation, as well as radiation induced immune cell death

Tumor RBE

Clinically applied values of (α/β)_x are a surrogate for all the mechanisms shown in figure 3 and are based on empirical evidence and data extracted from patient cohorts by fitting (α/β)_x dependent TCP (Tumor Control Probability) models to tumor growth or by comparing outcome of different fractionation regimens. Taking these (α/β)_x values [64–67], and assuming that RBE as a function of (α/β)_x follows the same trend as the one seen in CCS (Figure 2), leads to predicted RBE values shown in figure 4 for LET=2 (indicative of a typical lower boundary for the target), LET=4 (indicative of a typical average for the target), and LET=10 (indicative of a typical maximum for the target). Note that these absolute RBE values are model dependent and reflect predictions by the model shown in equation (2). However, the overall trend holds for all RBE models.

Figure 4:

Proton RBE at 1.8 Gy per fraction based on tumor (α/β)_x values extracted from clinical data [64–67] (excluding data with negative values) assuming RBE as a function of (α/β)_x) as deduced from clonogenic cell survival (equation (2) [4]) and LET estimates [21]. Upper: LET of 2 keV/μm representative for the proximal part of the tumor and likely a minimum conservative RBE; Middle: LET of 4 keV/μm representative for the distal part of the target; Lower: LET of 10 keV/μm as a maximum for the target.

The (α/β)_x from in vitro measurements are not one-to-one related to (α/β)_x extracted from clinical data. Whether in vitro measurements translate to clinically used (α/β)_x depends on the tumor histology and site, e.g., tumor types can have different (α/β)_x values depending on the tumor microenvironment (e.g., adenocarcinoma of the breast and prostate [65]). Table 1 compares RBE values shown in figure 4 with those from CCS in vitro.

Table 1:

Conservative tumor RBE values (Figure 3) in comparison to RBE values from data on clonogenic cell survival in vitro either measured directly or based on (α/β)_x and equation (1) (for references see figure 3 and text). As some laboratory experiments use 200–250 kVp X-rays as reference radiation, experimental RBE values were corrected by 1.1–1.15 to relate to 6-MV photons [9]. Also, data were corrected to correspond to ~1.8 Gy per fraction.

	Based on Clinical Data			Based on in vitro Clonogenic Cell Survival
	Range of (α/β)x [Gy]	Minimum target RBE according to equation (2)	Maximum target RBE according to equation (2)	Range of (α/β)x [Gy]	Minimum RBE reported or according to equation (1)	Maximum target RBE reported or according to equation (1)
Breast	2.2 – 4.6	1.08 – 1.18	1.31 – 1.60		< 0.93
Breast adenocarcinoma	4 – 4.4	1.14	1.48 – 1.50
CNS	1.8 – 12.5	1.07 – 1.19	1.27 – 1.63
CNS chordoma	2.4	1.18	1.59
CNS glioma	3.1 – 12.5	1.07 – 1.16	1.27 – 1.55	1.83	1.1 – 1.35	1.7 – 1.8
CNS meningioma	3.3 – 3.8	1.15 – 1.16	1.51 – 1.53
CNS vestibular schwannoma	1.8 – 2.4	1.18 – 1.19	1.59 – 1.63
H&N	0.9 – 30	1.03 – 1.23	1.14 – 1.71
H&N squamous cell carcinoma	0.9 – 29.2	1.03 – 1.23	1.14 – 1.71	1.46 – 47.5	0.9 – 1.4	> 2.1
H&N squamous cell carcinoma (larynx)	0 – 29.2	1.03 – 1.23	1.14 – 1.71	1.82 – 7.65	1.2 – 1.4	> 2.1
H&N squamous cell carcinoma (nasopharynx)	16	1.05	1.22
H&N squamous cell carcinoma (oropharynx)	6.5 – 10.3	1.08 – 1.11	1.30 – 1.40
H&N salivary gland				0.58 – 10.89	1.0 – 1.1
GI liver	13.4	1.06	1.46		< 1.15
GI esophagus	4.9	1.13	1.25		< 1.0 – 1.6
GI esophagus adenocarcinoma					< 0.6 – 1.0
GI esophagus squamous cell carcinoma					< 0.7 – 2.3
GI pancreas adenocarcinoma					0.7 – 2.4
GU prostate adenocarcinoma	0 – 20	1.03 – 1.24	1.14 – 1.74	5.8	1.0 – 1.85	> 2.0
GU colorectal				3.08 – 69.5	< 1.0
GU colon					< 1.21
GU cervix	10 – 52.6	1.01 – 1.08	1.08 – 1.31	6.9 – 7.1	1.07	> 1.28
GU cervix squamous cell carcinoma	26 – 52.6	1.01 – 1.03	1.08 – 1.15
GU rectum	2.7 – 11.1	1.07 – 1.17	1.29 – 1.57
GU bladder	13 – 24	1.04 – 1.06	1.16 – 1.26
Sarcoma	0.4	1.25	1.75	7.61	1.02
Lung (Non-small cell)	3.9 – 8.2	1.09 – 1.14	1.35 – 1.50	3.0 – 52.23	0.97 – 1.77	2.3
Pediatric rhabdomyosarcoma	2.8	1.17	1.56
Pediatric medulloblastoma				4	1.13

Breast:

Clinical data result in (α/β)_x of 2.2–4.6 Gy (4.0–4.4 Gy for adenocarcinoma). Using an empirical model, this corresponds to minimum RBE values for breast tumors between 1.08 and 1.18 in areas of low LET. For comparison, CCS results in the middle of an SOBP showed an RBE of 0.93 [15].

CNS - Head and Neck:

For CNS tumors the literature reports (α/β)_x between 1.8 and 12.5 Gy (2.4 Gy for chordoma, 3.1–12.5 Gy for glioma, 3.3–3.8 Gy for meningioma, and 1.8–2.4 Gy for vestibular schwannoma). For other head and neck tumors, (α/β)_x from negative values to 30 Gy were deduced from clinical data, the majority on squamous cell carcinoma. The large spread is reflected in the outliers in predicted RBE values seen in figure 4.

Indicative of potential minimum RBE in a tumor, in vitro data at low LET have been reported for various cell lines. For human salivary gland tumors cells, RBE values were reported as ~1.0 and ~1.1 at LET of ~2 [21]. For squamous cell carcinoma, RBE=1.4 (LET=2) [21] and RBE=1.0 (LET=2.1) [68] have been measured. RBE values for SQ20B derived from human epithelium tumors of the larynx translate into an RBE of ~1.2 (LET=2.4) [21] whereas Hep2 cells from a squamous carcinoma of the larynx showed RBE values of ~1.4 for LET of ~2.7 [21]. SCC25 derived from human epithelium tumors of the tongue showed an RBE of ~0.9 (LET=1.9) [21]. Three HPV-negative squamous cell carcinoma cell lines and 3 HPV-positive cell lines were studied with RBE values ranging from 1.15 to 1.33 in the center of an SOBP [69]. Using 3D cell cultures an RBE ~1.15 was measured for head neck squamous cell carcinoma at LET=3.7 [70]. With respect to expected maximum RBE values, measurements on SQ20B derived from human epithelium tumors of the larynx translate to RBE of ~1.25 (LET=7) and ~1.05 (LET=7.7) [21]. Furthermore, Hep2 cells from a squamous carcinoma of the larynx showed RBE values of ~2.1 for LET=8.7 and SCC25 derived from human epithelium tumors of the tongue showed RBE of ~2.1 (LET=8.9) [21]. In vitro measurements using glioma (U87) resulted in RBE of ~1.35 at LET=2.6 [71] and 1.1 at LET=2.26 (T98) [72]. Related to potential maximum target RBE values, the same studies report an RBE of ~1.7 at LET=13.4 (U87) and RBE =1.7–1.8 at ~7.3 (T98). For seven different glioblastoma cell lines RBE values between ~0.8 and ~1.3 were deduced [73] (no LET reported) as well as an RBE of >2.0, albeit at an LET of 6.2 [21].

Gastrointestinal:

Clinical data suggest (α/β)_x values of 13.4 Gy and 4.9 Gy for liver and esophagus, respectively. In the middle of an SOBP, RBE values for eight human HCC cell lines have been measured in vitro revealing large variations [17]. RBE values for 1.8 Gy/fraction were between 1 and 1.56 (at >2Gy; no (α/β)_x values reported). RBE values for human esophageal adenocarcinoma cell lines as well as squamous cell carcinoma cell lines were also measured in vitro [74]. For the two adenocarcinoma cell lines RBE values were not statistically different from 1.0 (at >2Gy; no (α/β)_x values reported). The difference between the squamous cell carcinoma cell lines was substantial with RBE values of 2.27 and 0.70. Using 3D cell cultures, large variations in RBE amongst six pancreatic duct adenocarcinoma cell lines were measured (RBE ~0.7 to ~2.4; no (α/β)_x reported) [75].

Genitourinary:

Data for adenocarcinoma of the prostate show (α/β)_x between 0 and 20 Gy from 63 different studies (50 of which showed values between 0 and 5 Gy). For cervix, (α/β)_x values deduced from clinical data range from 10 to 52.6 Gy, whereas (α/β)_x between 2.7 and 11.1 Gy were deduced for rectum, and 13–24 Gy for bladder.

Measured RBE values in vitro for the DU-145 prostate cancer cell line were low with ~1.0 at LET=2.9 [21]. Other studies using DU-145 found RBE values of ~1.83 at LET=1.9 [76] and ~1.85 at LET=1.9 [68], respectively. Representative for the distal part of the target, RBE values of ~1.98 at LET=4.1 [76] and ~2.08 at LET=4.5 [68] were reported. Furthermore, an RBE of 1.21 was deduced for colon carcinoma at 3.71 [21]. In vitro data for colorectal tumor cells report RBE values below <1.0 even for LET=7.7 and data for cervix predict an RBE of 1.07 and 1.28 at LET=2.4 and 6.9, respectively [21].

Sarcoma:

Clinical data from liposarcoma indicate a low (α/β)_x of 0.4 Gy suspecting a high RBE while in vitro data show an RBE of 1.02 for LET=2.56 [21].

Thoracic:

Large variation in (α/β)_x for NSCLC deduced from clinical data (3.9–8.2 Gy) as well as large variations in RBE in vitro have been shown. For A549 and H460 NSCLC cell lines, RBE values of 1.3 and 1.37 were reported for an LET of 2.5 [21]. Another study measured an RBE of 1.14 for A549 (LET=2.26) [72]. RBE values in the middle of an SOBP for SF=0.5 were measured for 17 different NSCLC cell lines in vitro with values between 0.97 and 1.77 [15]. In three of these cell lines, the increased RBE was correlated with alterations in the BRCA/Fanconi Anemia DNA repair pathway. In Calu-6, the data suggested the presence of a FANCD2 pathway defect leading to a previously unreported persistence of RAD51 foci after proton irradiation compared to the photon reference radiation. As for potential maximum RBE values in the target, A549 RBE of ~2.3 was reported for an LET of 10 [21] and an RBE of 1.45–2.31 for A549 was seen for LET=7.3 [72].

Pediatric:

Clinical data from rhabdomyosarcoma indicate an (α/β)_x of 2.8 Gy. As medulloblastoma are expected to have a higher (α/β)_x it was speculated that the RBE for medulloblastoma might be <1.1 potentially leading to failure in tumor control [77–79]. In a retrospective review of 109 patients with medulloblastoma treated with PT, no statistically significant correlation between areas of low LET (indicative of low RBE) and regions of recurrence was found, and patterns of failure did not differ from patients treated with photons [79]. Experimental in vitro data on cell survival of medulloblastoma cells report an RBE of 1.13 (LET=2.2) [21].

Normal tissue RBE

To ensure target coverage, the distal edge of a SOBP (in single field uniform dose delivery) extends beyond the tumor/CTV and therefore into normal tissue. While the beamlet distribution in IMPT causes a more complex plan optimization dependent LET distribution, elevated LET values would still occur mostly in the vicinity of the target contour (see Figure 1). Furthermore, considering typically lower (α/β)_x for normal tissues combined with lower doses compared to the target, one might expect larger RBE values when considering toxicities compared to tumor control. Tissues with high mitotic activity and rapid turnover and cycling stem/progenitor populations might respond early and regenerate more quickly after radiation damage [33]. Those would be expected to have a high α/β, if the α/β of the parenchyma alone is used to parameterize tissue response. In contrast, tissues with slow turnover might respond late and may rely more on proliferation and reprogramming of mature cells instead of cycling stem/progenitor populations [33]. Table 2 summarizes (α/β)_x values estimated from human data and figure 5 shows the RBE as a function of (α/β)_x from CCS. A figure similar to figure 4 for normal tissue is not meaningful. First, (α/β)_x from human data are based on fairly homogenous photon dose distributions in the region of interest. For more heterogeneous proton dose distributions the RBE has to be assigned on a voxel-by-voxel basis. Second, while in the case of tumor control, there is one direct pathway from cell survival to tumor cell reduction, normal tissue damage starts with damage to functional subunits leading to early and late toxicity (Figure 3). Furthermore, endothelial cells are organ specific, e.g., brain endothelial cells promote the blood brain barrier while capillary endothelial cells of the kidney allow blood filtration. The radiosensitivity of endothelial cells does depend on the organ with the turnover time of endothelial cells in an organ not necessarily related to the turnover time of the parenchyma [80].

Table 2.

Published values for (α/β)_x for selected early and late normal tissue endpoints. Shaded rows indicate endpoints with potentially high relative biological effectiveness due to low (α/β)_x

Tissue endpoints	(α/β)x [Gy]
Capillarics158	3
Connective tissue fibrosis158	2
Skin nccrosis/fibrosis159,160	1.9–2.3
Skin (dermatitis [erythema, desquamation])66,158	8.8–12.3
Skin desquamation66	18–35
Skin telangiectasis66,159	3.9–5.7
Ncrve (brachial plexopathy)160,161	2; < 5.3
Nerve (cauda equina)160	2–3
Breast (atrophy, fibrosis)158	2–3
Heart (peri card i ti s)158,160	2–3
Chiasm (loss of visión)158,160	2–3
Optic nerve (ncuropathy)158	2
Larynx (chronic edema, fibrosis)158,160	2–4
Larynx: cartilage necrosis and edema66,160	3.4–4.4
Salivary glands (xerostomia)158	3
Oral museositis158	10
Hairloss158	7
Oropharynx: late effects66	4.5
Brain/brainstcm necrosis160,162	2.1
Spinal cord myelopathy/necrosis66,160	2; <3.3
Livcr fibrosis158	1
Liver failure160	1.5
Esophagus (stricturc/pcrforation)160	3
Kidney ncphritis/ncphropathy158,160	2–3.5
Stomach (ulccration/pcrforation)158,160	4; 7–10
Colon (obstruction/ulccration/pcrforation)160	3.1–5
Rcctum (stcnosis/proctitis/necrosis)160	3.9
Rectum (chronic inflammation, ulcer)160	5
Small intestinc (obstruction/pcrforation)160	6–8.3
Bowel (stricture/perforation)66	2.2–8
Bladder160	3.4–6
Urinary bladder cystitis158	10
Ccrvical/thoracic/lumbar myelopathy158	2
Lung perfusion163	1.3
Lung fibrosis158,159	3–3.6; 4
Lung pneumonitis66,159,160	< 3.8; 4.4–6.9
Lung pneumonitis (early)66,158	> 8.8; 5
Rib fractures159,160	1.8–2.5
Bone (osteoradionecrosis)158	60

Figure 5:

Proton RBE for clonogenic cell survival as predicted by an empirical model [4]. Left: RBE as a function of (α/β)_x at 0.9 Gy (representative of 50% target dose for standard fractionation) for LET=2 keV/μm (dashed) and 10 keV/μm (solid). Right: RBE as a function of (α/β)_x at 1.8 Gy for LET=2 keV/μm (dashed) and 10 keV/μm (solid). The two LET values are representative for the entrance region and the distal fall-off of an SOBP field.

Breast:

In breast, no relation has been found between fibroblast radiosensitivity and the development of late normal tissue effects such as fibrosis, demonstrating the limitation of the target-cell hypothesis [81, 82]. Breast treatments often include a significant portion of the heart, which might lead to fibrosis and coronary heart disease, typically starting with microvascular injury inducing ischemia. Long term follow-up is needed to assess effects of potentially elevated RBE values. A randomized trial comparing proton and photon treatments is currently ongoing (NCT02603341).

A retrospective analysis of late-phase lung-density changes (indicative of asymptomatic fibrosis) for a small cohort of breast cancer patients irradiated to the chest wall showed that late-phase asymptomatic radiographic changes on follow-up CTs in the lung are associated with RBEs potentially even exceeding 3.0 [83]. When patients receive PT for breast cancer, ribs located in the distal region of the proton field may receive higher effective doses. An increased rib fracture rate of 7% was seen in PT for breast cancer. The LET in fractured areas was increased to ~6 suggesting that there is an end-of-range radiobiological effect and the RBE-weighted dose could be 20–30% higher than expected from RBE=1.1 [84]. Furthermore, in re-irradiation of locally recurrent breast cancer patients a higher rate of acute grade 3–4 skin toxicity and chest wall infections was seen with protons [85]. This seems likely a dosimetric rather than an RBE effect. Clonogenic cell survival suggest RBE values in skin fibroblasts between 1.4 and 2.2 at LET between 2 and 2.6 [21, 71, 86]. The RBE for skin keratinocytes at LET=2.1 was found to be 0.96 [68].

CNS - Head and Neck:

Brain injuries are caused by both vascular damage (vascular hypothesis), damage to glial cells (particularly oligodendrocytes and neurons; glial hypothesis) as well as interactions between the two [87, 88]. Injury to endothelial cells leads to vessel wall necrosis, loss of capillary vasculature and increasing permeability impacting the blood brain barrier which in turn may lead to vascular edema and hypoxia. This is a first step towards brain injuries [89]. Furthermore, regional differences in radiation sensitivity for white or gray matter or other areas of the brain might have implications with the more inhomogenous dose distribution offered by PT.

Neurocognitive toxicities are expected to be lower with protons due to the reduced integral dose [90]. Late contrast-enhancing brain lesions are indicative of a compromised blood-brain barrier. In six glioma patients, a logistic regression model presenting the probability of magnetic resonance (MR) image changes showed a better correlation when LET, as a variable, was included in addition to dose [91]. Numerous studies on temporal lobe necrosis after PT report incidence rates between 5.6 and 17.1% [92–95], i.e., slightly higher than comparable photon data [92]. Based on 110 low-grade glioma patients treated with PT, a voxel-based predictive model was developed for the occurrence of late MR contrast-enhancing lesions and was extrapolated to a patient-level risk model [96, 97]. A cross-validated voxel-level logistic regression to predict local risk for image change was used and predicted a threefold increased risk in the region around the ventricular system and an LET-dependent RBE of 1.22 (LET=2). A strong dependence on LET and an increased RBE for the ventricular proximity was seen even in regions of low LET. Another study on radiographic changes after PT for nasopharyngeal carcinoma estimated that an RBE~1.18 at D_1% [98]. In another study, the relationship between regions of elevated LET has been assessed for adult brain and head & neck patients and LET was not found to correlate with a risk of brain necrosis. It was concluded that RBE effects might be small compared to inter-patient variability of radiosensitivity and might be obscured by other confounding factors [99]. The complex relationships between vascular and glial mechanisms and their dose-volume effects likely contributes to this finding. Necrosis is likely correlated geometrically with the vascular structure and not only with direct cell kill to the brain parenchyma. Interestingly, there is a statistically significant difference in dose-response for brain necrosis depending on whether tumors are intra- or extracranial [100], emphasizing the importance of vascular effects.

A study in 100 glioma patients showed that the characteristics of pseudo-progression after PT differed from those observed after photon therapy [101]. Not only was the appearance different on MRI, also the timing was different as pseudo-progression after PT developed later compared to photons therapy (15 versus 3 months). Furthermore, there was an affinity for white matter with protons. The authors speculate that an increased RBE at the end of range might play a role.

Other toxicities of interest in head & neck cancer patients include xerostomia and dysphagia. These have been used as examples for model-based trials in PT [102]. Such efforts also aim at enriching patients for improvements in current photon based NTCP models and may allow future RBE estimates. Considering organ dose distribution is warranted. For instance, stem cells of parotid glands are predominantly located in the larger ducts [103].

Gastrointestinal:

Toxicities in the liver originate from damage to the parenchyma (hepatocytes) and to endothelial cells. Various imaging techniques have been used to study differences in radiation induced liver disease between protons and photon treatments. A study using SPECT/CT demonstrated that RBE might differ not only by patient but by baseline liver function [104]. Rib fracture also play a role in hepatocellular carcinoma [105]. A clinical trial comparing PT with IMRT for locally advanced esophageal cancers (NCT01512589) showed a significant reduction of toxicities using protons due to favorable dose distributions [106]. Such studies are not designed to deduce normal tissue RBE.

Genitourinary:

Subregions in the urethra and bladder have been shown to be more predictive for urinary toxicity than the dose to the whole bladder [107]. Late effects in the bladder wall or urethra arise from epithelial and microvascular alterations. A large trial comparing PT and IMRT for low or intermediate risk prostate cancer might ultimately provide RBE estimates (NCT01617161). Several large retrospective studies did not report significant differences in toxicities between the two modalities [108].

Sarcoma:

In bone sarcoma, radiation induced decrease in bone density can lead to fractures. Furthermore, a reduction in bone vascularity combined with impaired bone formation from irradiated osteoblasts affects fracture healing. Evidence suggests that bone fractures are impacted by elevated LET values with an estimate RBE at the end of range of 1.2–1.3 in ribs [84]. Fractures also occur in radiation treatment for soft tissue sarcoma but no comparative data have been published.

Thoracic:

The capillary endothelium is considered the initial site of pulmonary damage. Radiation pneumonitis leads to vascular permeability, hypoxia, and interstitial edema followed by fibrosis. The killing of fibroblasts leads to increased collagen production and density change. While vascular damage is generally more impactful in small vessels, a role for the major bronchi or associated large pulmonary vessels in the development of severe toxicity was suggested when small volumes of lung are irradiated with high doses [109]. Secondary to reduced vascular capacity in the irradiated region, vasculature in non-irradiated lung regions can be damaged due to enhanced pressure and overload [110]. These effects are important for integral dose considerations.

A retrospective analyses of late-phase lung-density changes in breast cancer patients irradiated to the chest wall showed that late-phase asymptomatic radiographic changes in the lung are associated with an elevated RBE [83]. In contrast, for the same endpoint, an RBE ~1.1 was deduced in a cohort of SBRT lung cancer patients indicating significantly lower RBE at higher doses per fraction. Differences in the time course of the inflammatory response after proton compared to photon SBRT were seen (inflammation occurred earlier in PT) [111]. A study using FDG-PET demonstrated that, despite significantly different dose distributions for IMRT and PT, the slope of the dose linear regression line associated with radiation pneumonitis did not differ and patients who developed pneumonitis had statistically significantly higher mean lung dose and higher slope regardless of treatment modality [112]. A randomized clinical trial comparing toxicities such as pneumonitis from PT versus IMRT (NCT00915005) for NSCLC did not find a benefit of PT [113]. The main reason might be the distribution of dose [114] and regional differences of radiation sensitivity [115]. In studies in NSCLC patients, non-cancer mortality was associated with irradiation of upper regions of the heart, including big vessels (e.g., the vena cava and coronary arteries) and the right atrium [116, 117]. The differences in dose distribution from photon and proton treatments might be more impactful than RBE effects.

Reported dose response data for CCS using human fibroblasts and epithelial cells derived from a lung fibroblast show high (α/β)_x, i.e., RBE values independent of dose of ~1.16 even at 7.7 [21].

Pediatric:

There is concern that elevated RBE for normal tissue injuries may lead to not only more but also more severe toxicities than expected based on dosimetric indices in pediatric patients [118]. Concerns focus mainly on the occurrence of brainstem necrosis in posterior fossa tumors [118–120]. Low necrosis rates of 1.1% [121] and 0.7% [122] were reported for pediatric low-grade glioma and 166 pediatric patients with CNS malignancies, respectively. Similarly, the incidence of brainstem injury in pediatric patients with posterior fossa tumors was reported as < 2% [123]. Another study reported symptomatic radiation-induced necrosis in pediatric brain tumor patients as 7% [124]. These numbers seem lower than what would be expected from photon treatments. Others reported a significantly higher incidence of radiographic changes (potentially necrosis) with protons at 31% [120] and, specific for pediatric ependymoma patients, 43% versus 17% in IMRT [125]. A study of 30 PT pediatric brain tumor patients reported 23% radiographic image changes [126]. A comprehensive analysis of childhood cancer survivors and CNS injuries showed a comparatively low necrosis risk of ~5% for pediatric PT [127]. Some studies demonstrate higher correlation of dose with image changes if an LET based correction is employed [126, 128]. The 5-year cumulative incidence of CNS injury for medulloblastoma patients receiving posterior fossa boost fields ranging out in the brainstem was 4% [129]. A total of 10 discrete areas of treatment change were contoured. Eight of these areas had higher LET values than the target, but with the average LET for these areas being 2.7 versus 2.4 in the target area, the difference is likely too small to indicate a correlation with RBE. The significant variations in numerous studies are due to small single institution cohorts and due to the definition of radiographic changes versus necrosis [118]. Presumably, areas of increased RBE had an impact only in those patients that were genetically prone to increased radiosensitivity.

The vasculature in children is likely more radiosensitive leading to vasculopathy [130]. A study in pediatric craniopharyngioma patients treated with PT showed a significant correlation with LET in vascular structures, while no correlation was found with dosimetric parameters alone [131]. On the other hand, in 644 pediatric PT patients with brain and skull base tumors, the 3-year cumulative rates of any vasculopathy and serious vasculopathy were 6.4% and 2.6%, respectively [130]. The data did not suggest rates that are higher in PT compared to photon therapy. The same conclusion can be drawn from a study on cerebral vasculopathy and microbleeds in pediatric brain tumor patients treated with PT [132, 133]. Such imaging studies may miss damage to small vessels or more long-term occurrences of vasculopathy.

Clinical comparison between the two modalities typically prevents distinction between effects due to RBE or due to dose distribution differences. For instance, a study on craniopharyngioma patients treated with protons and IMRT did find a difference in visual defects (52% versus 81% for proton and photon therapy, respectively) [134]. The same holds, for instance, for studies on intellectual outcomes demonstrating superiority of PT [135].

Conclusions

The combined effects of radiation action on tumor cells, tumor vasculature as well as systemic effects of radiation make it difficult to define tumor specific ‘clinical’ RBE values. Tumor dose prescriptions in PT are therefore based on a conservative RBE of 1.1. Potentially, slightly lower values could be encountered in proximal parts of the tumor volume for a given beam direction or for some hypo-fractionated regimen. While, for now, a conservative RBE is appropriate for prescription doses, more realistic RBE estimates should be applied when retrospectively analyzing outcome data. RBE values higher than 1.1 can be expected in the distal part of the tumor in SOBP fields if the relationship of RBE as a function of (α/β)_x as deduced from CCS holds for (α/β)_x deduced from clinical data. Furthermore, the average RBE across most tumors volume likely exceeds 1.1. The clinical relevance depends on the sigmoid dose-response curve. Ongoing large clinical trials offer the potential to deduce α/β for tumors treated with protons and might allow adjusting photon-based TCP models with an RBE defined by a shift in D₅₀. Additional studies on genomically characterized human cancer cell lines together with measurements of tumor control using human tumor cells implanted in immune-deficient animals would be valuable. Patient-derived organoids will play a big role as well [136]. While the current concept of RBE aims at population average treatment planning decisions, it is desirable to define biomarkers to identify patients that benefit most from PT. Efforts are ongoing to identify deficiencies in repair pathways. Functional biomarkers, such as foci size of DNA double-strand break markers signaling unrepaired clustered damage, may identify human tumors with increased sensitivity to proton radiation for treatment with radiation alone or in combination with targeted therapies [17, 70, 75, 137]. Radiation therapy is often used in combination with chemotherapies, targeted agents and immune therapies, all of which influence the RBE. For the latter, the impact of radiation dose distributions on the lymphatic system and lymphocytes as well as on immune-related signaling by normal tissues around the tumor needs to be better understood.

An RBE of 1.1 is currently also applied for normal tissue constraints. Even though one would expect smaller variations in the underlying cell population for normal tissues as compared to tumors, the mechanisms leading to normal tissue toxicities are more complex due to the variety of toxicities but also due to the potentially higher impact of vascular damage as compared to cell death in the parenchyma. The lower integral dose in PT may not translate into a toxicity advantage depending on how the dose is distributed and depending on normal tissue RBE [108]. The RBE needs to be applied inhomogenously according to the dose distribution and the distribution of functional subunits. Moreover, it is unclear if current NTCP models are applicable because current models are based on fits to clinical data from dosimetrically more homogeneous photon treatments. In this context, dose-response relationships should not be solely analyzed based on organ contours but on sub-regions or even voxel-based [138]. Personalized normal tissue radiosensitivity might be assessed via transcriptome response [139, 140]. Furthermore, the clinical data on normal tissue toxicities reviewed above have limitations not only because of patient variability but also in terms of institutional bias from treatment planning, delivery, or outcome assessment. There have been many planning studies on RBE effects in PT which are not reviewed here [141–152]. Such studies apply empirical models based on CCS and are valuable to indicate general trends and potential impact of RBE variations in normal tissues. Caution is warranted when interpreting these RBE values quantitatively.

This review focuses on biological and clinical aspects of RBE. There have been other reviews focusing more on the physics of proton biology, i.e., the specific impact and calculation of LET as well as the shaping of LET distributions in treatment optimization [2, 153]. With the difficulty to define reliable RBE values for normal tissue complications, a safe strategy might be to maintain current dose constraints but reduce LET in regions of interest via LET based plan optimization [153–157].