Spectral complexity of 5-ALA induced PpIX fluorescence in guided surgery: a clinical study towards the discrimination of healthy tissue and margin boundaries in high and low grade gliomas

Gliomas are diffuse and hard to cure brain tumors. A major reason for their aggressive behavior is their property to infiltrate the brain. The gross appearance of the infiltrative component is comparable to normal brain, constituting an obstacle to extended surgical resection. 5-ALA induced PpIX fluorescence measurements enable gains in sensitivity to detect infiltrated cells, but still lack sensitivity to get accurate discrimination between the tumor margin and healthy tissue. In this fluorescence spectroscopic study, we assume that two states of PpIX contribute to total fluorescence to get better discrimination of healthy tissues against tumor margins. We reveal that fluorescence in low-density margins of high-grade gliomas or in low-grade gliomas is mainly influenced by the second state of PpIX centered at 620 nm. We thus conclude that consideration of the contributions of both states to total fluorescence can help to improve fluorescence-guided resection of gliomas by discriminating healthy tissues from tumor margins. © 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement


Introduction
Diffuse gliomas account for more than fifty percent of primitive brain tumors and are currently hardly curable.All subtypes share the same highly infiltrative behavior of individual tumor cells.However surrounding infiltrated tissue often resemble normal tissues.The world health organization (WHO) classifies gliomas in 4 grades [1], but most studies commonly consider two separate groups having different biological, molecular and tissue properties: High Grade Gliomas (HGG) and Low Grade Gliomas (LGG).HGG are mainly malignant tumors (grades III and IV of WHO classification), while LGG are benign tumors (grades I and II).In both groups, infiltrative tumor cells are still difficult to identify during surgery.Studies have shown that in 85% cases, recurrences of HGG are localized less than 2 centimeters from the first tumor [2].Improving the extent of resection prooved to prevent recurrence and improve life quality as well as life expectancy [3][4][5].
Pre-operative MRI combined with neuro-navigation is currently used to localize surgical tools and tumor cells in the operating theater [6,7] but it shows strong limitations.Indeed, MRI lacks sensitivity to detect tumor margins [8,9].Furthermore, the brain shift can reach up to 3 cm in comparison with pre-operative MR images [10].In order to prevent any localization error due to the brain shift, intraoperative MRI has been suggested.Nonetheless, intraoperative MRI complicates the surgery and has the inherent lacks of sensitivity of MRI to detect tumor margins, which makes it barely used.
As a complementary method to pre-operative MRI, fluorescence microscopy has shown its relevance in neuro-oncology [11][12][13][14][15] and 5-aminolevulonic acid (5-ALA) induced fluorescence of protoporphyrin IX (PpIX) is currently used through surgical microscopes [16].PpIX takes part in the biosynthesis of heme and ingestion of 5-ALA, precursor in heme's biosynthesis, leads to temporary enhancement of PpIX concentration in gliomas [17].PpIX absorbs light around 405 nm and emits a reddish fluorescence with a main peak centered at around 634 nm.PpIX fluorescence-guided resection of gliomas through surgical microscopes enabled to gain 6 months life expectancy for HGG [18].This technique is the actual clinical standard for PpIX-based surgical assistance.It is also investigated for intracranial stereotactic biopsies [19,20].However, its sensitivity is still limited when applied to low density infiltrative parts of HGG [21,22] or to LGG [23].Some studies also emphasize that the link between the subjective fluorescence intensity scale chosen by the surgeon and the pathological status of tissues is still to be clarified [24,25].
To overcome sensitivity issues, various 5-ALA induce PpIX fluorescence spectroscopy methods have been proposed.These studies could be classified into two main approaches, either assessing quantitative concentration of PpIX, or qualitative or semi-quantitative biomarkers related to PpIX fluorescence.Quantification of PpIX concentration has been proposed with combined measurements of fluorescence spectroscopy and reflectance to correct tissue distortions [26,27].This technique is more sensitive than fluorescence microscopy for HGG [28].It has also been used in LGG but its sensitivity is yet only up to 45% [29].Some works proposed normalization procedures or other qualitative or semiquantitative biomarkers to increase the robustness of PpIX fluorescence measurement [19,[30][31][32].It should be pointed out that both approaches are relevant because all these measurements are intermediary steps toward the final aim of PpIX fluorescence measurements, which is the classification of measurements into relevant pathological status.
Quantification techniques rely on the assumption that the emitted PpIX fluorescence intensity is proportional to the concentration of PpIX.This is supported by the link between the fluorescence emitted intensity and the tumor cellular density [17,33,34].However, some studies showed more complex links between the PpIX concentration and the shape of fluorescence spectra.A study on biopsies from 5 patients suggests a wavelength shift of the peak intensity of the emitted spectrum in a tumor and nontumor regions [31].Another study on biopsies raised the presence of two states of PpIX with different fluorescence spectra, peaking at 634 nm and also at 620 nm, in HGG as well as in LGG [32].
The presence of the second peak of fluorescence of PpIX at 620 nm is known in solution and closely linked with the chemical microenvironment [35][36][37][38].This peak has been observed in tissues [32,[39][40][41] or in cell culture [42,43].But its origin in vivo is still an open issue.Some works support the assumption that the origin of the peak at 620 nm in vivo is a different aggregate of PpIX [32,42].Other works [39,43] explained it by precursors of uroporphyrins or coproporphyrins.
The present study aims at investigating the presence of the second peak of fluorescence of PpIX at 620 in non-extracted and freshly extracted living brain tissues, and its correlation with the pathological status of tissues.To do so, we conducted a feasibility clinical trial on 10 patients.An intra-operative fluorescence spectroscopy system has been developed to measure fluorescence spectra of tissues with different pathological status, from healthy tissues to tumor, in HGG and LGG.The complexity of the fluorescence spectra has been investigated by a fitting process considering the two states of PpIX, peaking at 634 nm and also at 620 nm.This led to the definition of new biomarkers of PpIX fluorescence emission, which rely on the concentration of both states of PpIX.These results suggest that these biomarkers could increase the sensitivity of the 5-ALA induced PpIX fluorescence in low density margins and also to the dis and also in LG

Patient s
The study wa 10 patients w low-grade glio All exper ethics commi written conse diagnosis of judged suitab Exclusion crit or hypersens history of liv aminotransfer limit at any ti patient inabil and serious as and/ or T2navigational g

Spectros
The develope the setup that system, a box to excite the through a lig Maximum (M M0600T, sedi The chara phantoms in standardizatio fluorescence s The inner close to curre always at the probe, excitin with the phot hiatric illness.P al magnetic r rawing of the setup o be user frien of a personal ectrical and opt and collect flu iode (LED) ce rlabs).Light robe developed f the system udy [37].We u by [44]  cleaved so that excited tissue area and emitting tissue area are the same.Light goes through a low pass filter with a rejection band between 450 nm and 700 nm.This led to an output irradiance of 30 W/m 2 .Light was collected through the same probe, with a detection fiber (600 µm diameter) that is placed in the center.Light is then sent to a high pass filter with an in-band transmission between 485 nm and 1200 nm.The filtered light was finally injected into a spectrometer (Maya2000, Ocean optics).
Sterilization of the probe was performed by autoclave process, as recommended by French laws.This implies a steam cycle reaching 134°C and a relative variation of pressure of + 2 bars and −1 bar.Prion risk was tackled with a manual cleaning with an ALKA100 solution (Alkapharm).

Surgical procedure and data acquisition
Patients were given an oral dose of 20 mg/kg of body weight of 5 amino-levulinic acid (Gliolan; Medac GmBH) approximately 3 hours prior to the induction of anesthesia.The patient's head was prepared and registered using a StealthStation Treon image-guidance system (Medtronic) following standard practice.A Zeiss OPMI Pentero surgical microscope (Carl Zeiss Surgical GmbH) modified for fluorescence guidance with a 400-410 nm wavelength source for excitation and a 620-710 nm bandpass filter to record fluorescence emissions on a sensitive 3-chip CCD camera was also coregistered with the surgical field.
For each patient, standard surgical procedure started in order to expose the expected tissue.When asked by the surgeon, surgical procedure was stopped and fluorescence study started.Visual fluorescence was evaluated through surgical microscope.Then, the probe was placed on the tissue to be analyzed, light of the room were lowered and in vivo spectroscopic measurements started.Each acquisition was composed of 200 ms of duration with the LED turned on followed by the same duration with the LED turned off to get rid of ambient light coming from the operating room.For each measurement, 12 acquisitions were led, giving a total acquisition time of 4.8 s.The tissue was then removed from the brain and a new measurement was performed on the biopsy (named "ex vivo measurement" thereafter) before sending it for histopathological analysis.These fluorescence measurements were performed 6 to 8 times per patient at various stages of surgery.Since the goal of this study is to address the critical stage of minimum visible residual tumor, most of the samples were chosen at the sensitivity limits of the surgical microscope and MR images or beyond them.
In total, 174 measurements were performed and 143 were kept in this analysis.31 measurements were removed either because they showed no detected fluorescence signal, or due to heterogeneous pathological status, or due to failure to comply with part of the procedure.

Histopathology
Histopathological analysis was performed on formalin fixed paraffin embedded biopsy tissue specimens processed for HES staining.Each H & E stained tissue section was assessed for the presence of tumor cells, necrosis, mitotic activity, nuclear atypia, microvascular proliferation and reactive astrocytosis.Molecular criteria were also assessed.Biopsy specimens were then classified according to WHO histopathological and molecular criteria [1].Samples were sorted into 5 groups according to their histopathological appearance: HGG solid part, HGG margin, HGG margin of low density, LGG and healthy tissue.Samples classified as "heterogeneous" i.e containing morphological aspects of 2 different categories by the pathologist were removed from the study.

Data pre-processing
For each measurement (in vivo or ex vivo) of each analyzed tissue, emission spectra was obtained as follow: i) each 'off' spectrum was subtracted from the previous 'on' spectrum to get 12 fluorescence spectra ii) the 12 spectra were summed up to obtain 1 emitted spectrum

Signal to
We addressed status.Howev due to differ measure two divided by th 930-1030 nm

Results
The lipofusci whereas the p vivo and ex vi

Contribu
One objective Thus, Fig. t that can 4 is really e order of magnitude).The second one is the HGG low density margins and LGG where the mean contribution of the state 620 is slightly higher than the one of the state 634.The third group is the healthy tissues where the mean contributions of each state are almost equal and the total contribution of the two states is lower than the other groups.Table 1 displays the numerical values featured in Fig. 3 as well as PpIX fluorescence intensity, evaluated by integrating the fluorescence fitted curve between 600 and 650 nm, and central wavelength of the fluorescence spectrum in the same range.The central wavelength was defined to be the average wavelength of the interval representing more than 80% of the maximum's fluorescence intensity.1 confirms the decrease of total fluorescence intensity when the concentration of tumor cells decreases for HGG, with one to two orders of magnitude of fluorescence intensity higher in solid part of HGG (~2300) than in the others classes (between 30 and 250).Comparing the central wavelength, a shift to lower wavelengths (from 635 nm to 629 nm) appears with decreasing concentration of tumor cells for HGG, as exposed by [31].The central wavelength value appears to be lower for LGG and HGG low density margins than the others.Finally, the contribution of State 634 decreases from 59 to 0.43 as the concentration of tumor cells decreases, while the contribution of State 620 increases from 0.31 to 0.72, apart for healthy tissues where contributions of the two states are low.

Ratio of the two contributions
The ratio of the two contributions (α 620 / α 634 ) is drawn on Fig. 4 for each state.This ratio was computed in order to compare it with previous results on biopsies [32].This ratio tends to 0 in solid parts of HGG and increases as the density of tumor cell decreases.In HGG low density margin and LGG the ratio is higher than 1 and reach around 1.5 in LGG, showing the preponderance of State 620 in LGG.However, this ratio decrease and tends to 1 in healthy tissues.

Signal to
Results above biomarkers ar (Fig. 5) and f separated HG classification classification already define  However, the number of measurements is too low to investigate the statistical relevance of this assumption.This remark is even more striking with the patients of LGG of the Fig. 6.Indeed, the LGG samples have SNR values clearly higher than healthy tissues, even though, as for HGG, no statistical tests support this assumption.

Discussion
This study confirms the predominance in vivo of the commonly studied State 634 of PpIX in the tumor cells high-density regions of HGG (solid part and margin classes) where the ratio (α 620 / α 634 ) is close to 0. It is not the case in low-density tumor cells regions (low-density margin of HGG and LGG) where this ratio is higher than 1.

Low PpIX fluorescence intensity in low-density tumor cells regions
The contribution of state 634 is 10 to 100 times higher in high-density tumor cells regions as compared to healthy or low density tumor cells regions.This should be related to the actual sensitivity issues of the 5-ALA induced PpIX fluorescence surgical microscopy in the low density margin of HGG [21,22] and in LGG [23].Surgical microscopy fluorescence intensity was also assessed in this study.Data, not shown here, confirm that samples with fluorescence detected with the microscope are mainly localized in high-density tumor cells regions where the contribution of State 634 widely dominates.Quantification of PpIX concentrations is more sensitive than fluorescence microscopy but also shows sensitivity issues in low density tumor cells regions in HGG [28] or LGG [29].These techniques rely on the assumption that the emitted PpIX fluorescence intensity is proportional to the concentration of PpIX.This is supported by the link between the fluorescence emitted intensity and the tumor cellular density [33,34].However, some studies show that the microenvironment could have an important role in the intensity of fluorescence.Indeed, a strong 5-ALA induced fluorescence intensity has been observed in a patient showing only inflammatory cells without tumor cells [47].Furthermore, PpIX accumulation is known to be sensitive to gene expression [48] or metabolic pathways [49].Our study shows the predominance of State 620 of PpIX in these low-density tumor cells regions.It also shows that the contribution of State 620 is twice higher in these low-density tumor cells regions as compared to healthy and high-density tumor cells regions.It should be pointed that the fluorescence quantum yield of the two aggregates of PpIX are very different, the one of State 634 being far higher than the one of State 620 [36].Then, a decrease in fluorescence intensity might not only reveal a decrease in PpIX concentration but also a change of PpIX aggregate, in favor of State 620 [37].Our results suggest that the decrease of PpIX fluorescence intensity in low-density tumor cells regions [21][22][23] attributed mainly to low PpIX concentrations could also be linked to a different aggregation of PpIX in these regions, due to a microenvironment favoring the PpIX State 620.The similarities in the PpIX fluorescence behaviors in LGG and HGG low density margins could be explained by biological processes rather close from each other: presence of infiltrative but not joined tumor cells without blood brain barrier breakage around.

Towards a discrimination between healthy tissues and margins
This study shows the low contribution of both states of PpIX in healthy tissues, which is consistent with previous study showing low PpIX concentration [17,27,50].The contribution of State 634 is constant in healthy tissues and in LGG, and increases of only 37% in low density margin of HGG.This is consistent with the actual sensitivity issues of the 5-ALA induced PpIX fluorescence surgical microscopy in HGG and LGG [21][22][23].However, the contribution of State 620 is 80% higher in LGG and 95% higher in low density margins of HGG than in healthy tissues.This opens new insights in the use of this technique during HGG and LGG neurosurgery to discriminate healthy tissues from margins.

Blue shift of PpIX fluorescence spectrum in low-density tumor cells regions
Table 1 shows a shift of the central wavelength of emitted spectra towards short wavelengths when the density of tumor cells decreases.This shift is almost of 6 nm for low-density tumor cell margins of HGG and even 8 nm in LGG.A study on biopsies also suggests a shift in the same direction of the peak intensity of the emitted spectrum when the density of tumor cells decreases [31], but with a lower magnitude of around 2 nm ± 2 nm.However, in this latter study the wavelength shift was investigated in a margin sample.Then, it is consistent with the shift from high to low density in our study, which was 4 nm.The presence of a second peak of fluorescence at 620 nm, which produces this shift, has been shown in tissues [32,[39][40][41] or in cell culture [42,43].It is known that such a shift can be induced by a change in PpIX aggregation, which modifies the proportions in State 634 and State 620 of PpIX [35][36][37][38].Hence, the blue shift measured in this study in low-density cell tumor region is another indication that HGG margins and LGG could produce a different aggregation of PpIX, due to a microenvironment favoring State 620 of PpIX.

Ratio of PpIX contributions
Our previous study explored the ratio of the two contributions of PpIX in HGG and LGG, and showed significant difference between the solid part and margins of HGG [32].The present study explores this ratio with more precision according to the tumor cell density in tissues.In HGG, the ratio in high density margins is very close to the ratio in the solid part.The significant difference is between the high-density to low-density tumor cell regions, where the ratio reaches values higher than 1.It seems that the relevance of the ratio to discriminate tissues status would rather be in the high to low density transition, instead of solid part to margin transition as was suggested earlier [32].In LGG, the ratio is very close to the one observed on low-density tumor cells region of HGG.This confirms the similar PpIX fluorescence behavior that seems to appear in LGG and low-density tumor cell region of HGG, as discussed above.In healthy tissues, the ratio is close to 1.Its relevance to discriminate healthy tissues from margins has to be investigated further.

Origin of the fluorescence peak at 620 nm in vivo
The presence of the second peak of fluorescence of PpIX at 620 nm is known in solution and closely linked with the chemical microenvironment [35][36][37][38].In particular, PpIX is known to be pH related [32].The state 620 of PpIX is favored by high pH and the state 634 is favored by low pH.Our results show that the fluorescence peak at 620 nm is preponderant in lowdensity tumor cells regions; whereas the fluorescence peak at 634 nm is preponderant in highdensity tumor cells regions.This can be related to higher acidity in high-density tumor cells regions of HGG as compared to low-density tumor cells regions of HGG and in LGG since it is known that the tumor tissues acidity is correlated to the grade of glioma because of high glycolytic activity [18].This tends to suggests that the peak at 620 nm is due to a different aggregate of PpIX.However, this study did not retrieve a preponderance of either states of PpIX in healthy tissues.In particular, the contribution of state 620 is not higher than the one of state 634, which would be consistent with the hypothesis of a high pH stated above.Moreover, PpIX is lipophilic; in vivo, its distribution is mainly intra-mitochondrial and PpIX is in contact with the lipid membrane of mitochondria, so its microenvironment is quite complex.Furthermore, some results indicate that in ALA-induced and tumor conditions PpIX could spread outside the mitochondria and be partly sensitive to cytosolic microenvironment [51]-[53], or even to extracellular microenvironment [51].Some works support the assumption that the origin of the peak at 620 nm in vivo is a different aggregate of PpIX [32,42].Other works [39,43] explained it by precursors of uroporphyrins or coproporphyrins.This point is still controversial and should be explored, but it is out of the scope of this study.

Towards classification biomarkers in HGG and LGG
We investigated the relevance of simple biomarkers, the SNR, directly extracted from this data set with the idea of going towards a classification process into relevant pathological status.It should be emphasized that this is only preliminary work to evaluate the relevance of this way.We separated HGG and LGG, but this process does not imply a classification of LGG versus HGG.The grade of glioma is evaluated before the surgery by other means.We used the same four classes for HGG and two classes for LGG as the one used before in this study.This classification is rather subjective and dependent on the histopathologist.Indeed, we sampled a continuous process (density of tumor cells) that can induce a bias.In HGG, Fig. 5 shows the solid part along the state 634 axis and clearly distincts it from the other groups.The low density and high density margins groups are also distinct from each other, with a separation line which arises from the equal contributions line.Finally, the healthy tissues group seems somehow mixed with the margins groups, but the low contributions are apparent.In LGG, Fig. 6 shows a clear distinction between LGG and healthy tissue.Some studies emphasize the fact that the link between the subjective fluorescence intensity scale chosen by the surgeon in fluorescence microscopy and the pathological status of tissues is still to be clarified [24,25].Furthermore, classification based on fluorescence of PpIX is used in LGG but show lower sensitivity and specificity [29] than in HGG.Then our results seems rather different, showing that classification based on state 634 and state 620 could highly improve the intraoperative classification efficiency.

Model of fluorescence emission
The fluorescence emission model used in the parametric data analysis has been discussed previously in detail [32,37].However further improvements are added in these study with the fitting of lipofuscin.Lipofuscin concentration is known to be localized around the nucleus of neuronal cells, and its fluorescence emission spectrum shape is known to be linked with aging, oxidation degree and cell stemness degree [46].The relevance of measuring lipofuscin fluorescence in gliomas can be found in [49] to avoid bias in quantification of the PpIX concentration.Thaw et al [51] presents that fluorescence intensity of Lipofuscin is higher in glial than in glioma cells.As the presence of glioma cells makes glial cell density decrease, the Lipofuscin fluorescence intensity is lower in the solid part or high density margins of the tumor than in heathy tissues and low density margins of the tumor.Data not shown here demonstrate a linear relationship between lipofuscin fitting coefficient and the age of the patient.The oxidation degree has been observed as well since the contribution of lipofuscin appeared to be higher in ex vivo measurements than in in vivo ones for an identical sample.Thus lipofuscin fluorescence intensity cannot be neglected in some cases.As the shape of lipofuscin fluorescence is a Gaussian curve and the central wavelength is around 585 nm [11], the contribution of the state 620 is overestimated compared to the one of the state 634 if lipofuscin fluorescence is not extracted.Thus, lipofuscin fluorescence must be extracted to avoid bias in the contributions of both states and between the samples whatever the patient age and the measurement condition (ex vivo or in vivo).

Conclusion
In this study, we added the contribution of a second state of PpIX (named State 620) to the well-known and commonly used reference spectra of PpIX (named State 634) to analyze fluorescence spectra of glioma.The goal of this intraoperative clinical trial was to investigate the low sensitivity of fluorescence measurements in low density margins of HGG and LGG and to confirm the power of the second peak of PpIX to help discriminate glioma against healthy tissues.To do so, a linear fitting process was implemented and new biomarkers appeared to be of great interested.Those biomarkers, α 620 and α 634 first and then SNR 620 and SNR 634 are quick easy to get intraoperatively and could thus help improve fluorescence guided resection of glioma.Indeed, those results confirm that the contribution of State 634 dominates in HGG and their high density margins.What is more, they reveal that the contribution of State 620 is higher than the one of State 634 in low density margins of HGG and in LGG.Blue shift of the central wavelength when the density of tumor cells decreases confirms the weight of State 620 in low density margins.Those results could help understand the lack of sensitivity of current techniques, looking for State 634 where the dominant state is State 620.Based on those two states, the next step is to increase data set and investigated new supervised and unsupervised classification methods to make the discrimination even more robust.Using both states, the discrimination between healthy tissues and regions with a low density of tumor cells appears to be feasible for HGG and LGG as well, offering new opportunities to increase the sensitivity of fluorescence measurements and thus improve the extent of resection and reduce recurrence.
640 this figure is are the two st is defined by where α i inc concentration system geome and A lipo , λ lipo Fig. 2 contri spectr (dotte 55 yea The refere described in references we spectrometer depending on the reference State i (620 or f the Gaussian c (a) and fit of PpI fit (solid red cu ofuscin (solid gree ow density margin ere measured r total fluoresc 34 (λ) for comp e emitted spec re left free to s Due to fluore y

Figure 5
Figure 5 confirms that HGG solid part has an overwhelming contribution of SNR 634 and almost no contribution of SNR 620 .Most of the HGG margins are above the equal contribution line (dotted line) showing a preponderance of SNR 634 .Whereas the HGG low density margins are below this line, showing a preponderance of SNR 620 .Healthy tissues from the HGG patients are more unequally spread but seem to concentrate in the low SNR of both states region.This figure indicates clearly a different behavior for these four different pathological classes.However, the number of measurements is too low to investigate the statistical relevance of this assumption.This remark is even more striking with the patients of LGG of the Fig.6.Indeed, the LGG samples have SNR values clearly higher than healthy tissues, even though, as for HGG, no statistical tests support this assumption.