Tumors

The clinical value of microarrays was shown by studies in breast cancer,^6,7 lymphoma,⁸ lung adenocarcinoma,⁹ glioma,¹⁰ and others that showed tumors could be subdivided into groups based on their gene expression profiles—and that these subdivisions have clinical relevance, since the different groups have different prognoses (Fig. 5-1). In addition, the prognostic potential of microarrays was shown by van’t Veer and colleagues,¹¹ who defined a 70-gene signature that predicted the chance of distant metastases in young women with breast cancer. Several other signatures with the same utility have been found in subsequent studies, often containing different genes.^12,13 Of note is that these different signatures usually select the same patients as being at high or low risk, suggesting that many different gene sets may have similar prognostic potential for the same clinical situation and may represent different genes on common deregulated pathways.^14–16 In addition to distant metastases, it also appears to be possible to predict the chance of developing regional lymph node metastases in sites such as head and neck, breast, and others.^17–19

FIGURE 5-1 • Examples of the clinical value of gene expression profiling by microarrays. (A) Study by Sorlie and colleagues⁷ on breast tumors showing that clusters defined by differences in expression patterns have significantly different outcomes. (B) Study by Garber et al.⁹ showing a similar result for lung tumors.

(Redrawn from Sorlie T, Tibshirani R, Parker J, et al: Repeated observation of breast tumor subtypes in independent gene expression data sets, Proc Natl Acad Sci USA 100:8418–8423, 2003; Garber ME, Troyanskaya OG, Schluens K, et al: Diversity of gene expression in adenocarcinoma of the lung, Proc Natl Acad Sci USA 98:13784–13789, 2001.)

It is useful at this stage to distinguish between the terms prognostic and predictive. Several gene signatures have been described that appear to discriminate between good and poor outcome, but in a variety of disease sites and in patients given a variety of different treatments. Such signatures can be described as prognostic and probably reflect degree of malignancy. Examples are the “wound” signature,²⁰ hypoxic signatures,²¹ genetic instability signatures,²² and stem cell signatures.²³ By contrast, a signature that discriminates between good and bad responders to a specific treatment can be described as predictive. Such signatures can help in treatment selection, whereas prognostic signatures are inherently less useful. A predictive signature can also, in principle, help unravel causes of resistance, leading to potential new intervention strategies.

To date, very few predictive signatures have been described. Chung and colleagues²⁴ defined a 75-gene high risk signature for patients with head and neck cancer treated with primary surgery followed by radiation and/or chemotherapy. Pramana and colleagues²⁵ subsequently showed that this signature also predicted locoregional control in an independent series of head and neck cancer patients treated with radiotherapy and concomitant cisplatin (Fig. 5-2). However, the specificity of this signature for radiotherapy, with or without chemotherapy, remains to be determined. Nuyten and colleagues²⁶ described a signature for predicting local recurrence after breast conserving therapy (local excision plus radiotherapy). Signatures have also been described that predict tamoxifen resistance in breast cancer.^27,28 Validated signatures predicting response to radiotherapy alone in cancer patients have not yet been reported.

FIGURE 5-2 • Chung et al.²⁴ defined an expression signature comprising 75 genes associated with high risk of relapse in head and neck cancer treated with primary surgery with or without subsequent treatment with radiation and/or chemotherapy. The signature significantly distinguished good and poor prognosis groups in an initial group of 28 patients (A) and in a subsequent group of 60 patients (B). Pramana et al.²⁵ found this signature to predict locoregional control after concurrent radiotherapy and cisplatin in an independent group of 70 head and neck tumors (C).

(Redrawn from Chung CH, Parker JS, Ely K, et al: Gene expression profiles identify epithelial-to-mesenchymal transition and activation of nuclear factor–kappaB signaling as characteristics of a high-risk head and neck squamous cell carcinoma, Cancer Res 66:8210–8218, 2006; Pramana J, Van den Brekel MW, van Velthuysen ML, et al: Gene expression profiling to predict outcome after chemoradiation in head and neck cancer, Int J Radiat Oncol Biol Phys 69:1544–1552, 2007.)

MicroRNAs (miRNA) also have been shown to have predictive or prognostic potential. These are small 18- to 22-nucleotide single-stranded RNAs. They are transcribed from genomic DNA like messenger RNA (mRNA), but do not code for proteins. Instead, they bind to partially complementary sequences on target messenger RNAs, thereby inhibiting translation and sometimes causing mRNA degradation. It is estimated that the expression of up to half of all genes are regulated by miRNAs. It is therefore not surprising that they have been shown to be involved in carcinogenesis and treatment response. Reports are now appearing of the predictive potential of miRNAs,^29,30 which may ultimately prove to be at least as powerful as mRNA profiling for prediction (Fig. 5-3). To date, no study testing miRNA as a predictor of outcome after radiotherapy has been reported, although specific miRNAs have been shown to be induced by hypoxia and to correlate with outcome in breast cancer patients treated by surgery and adjuvant therapies.²⁹

FIGURE 5-3 • The prognostic potential of microRNAs (miRNA). Yu et al.³⁰ showed that a 5-gene miRNA signature distinguished good and poor outcome groups in non–small cell lung cancer in a training group of 56 patients (A) and in a test group of 56 patients (B). Camps et al.²⁹ showed that a single miRNA, hsa-miR-210, which is induced under hypoxia, could distinguish good and poor outcome in 219 breast cancer patients. Camps et al. showed that a single miRNA, hsa-miR-210, which is induced under hypoxia, could distinguish good and poor outcome in 219 breast cancer patients (C).

(Redrawn from Yu SL, Chen HY, Chang GC, et al: MicroRNA signature predicts survival and relapse in lung cancer, Cancer Cell13:48-57, 2008, and Camps C, Buffa FM, Colella S, et al: hsa-miR-210 Is induced by hypoxia and is an independent prognostic factor in breast cancer, Clin Cancer Res 14:1340-1348, 2008.)

Methylation of DNA in the so-called CpG islands in gene promoter regions can silence expression of those genes. Conversely, demethylation can activate transcription. Deregulation of methylation can thus lead to silencing of tumor suppressors or activation of oncogenes. Deregulation of genes involved in drug metabolism, the DNA damage response, and others can affect response to chemotherapy and radiotherapy. The most consistent result concerning prediction is in glioblastoma, where methylation of the O-6-methylguanine-DNA methyltransferase (MGMT) promoter is associated with improved survival in patients treated with radiotherapy plus alkylating agents, particularly temozolamide.^31,32 The MGMT enzyme can reverse drug-induced DNA alkylation, thus reversing its cytotoxic effects. Gene methylation will reduce expression of this DNA repair gene, leading to increased tumor responses. Studies in other tumors have also shown correlations between methylation status of specific genes and outcome in colorectal cancer,³³ neuroblastoma,³⁴ and others. No clinical studies on radiotherapy alone have reported on methylation status. Several methods have been developed to measure DNA methylation, some of them approaching a genome-wide scale, and are now being increasingly applied to test their predictive potential.

Comparative genomic hybridization (CGH) measures copy number variations (CNV; amplifications and deletions) in regions of genomic DNA, currently at a 5 to 10 kilobase resolution. This method allows characterization of recurrent chromosome changes in tumors, which can pinpoint relevant known or novel oncogenes and suppressor genes. In addition, CNVs can be correlated with outcome to define CNV predictors, analogous to gene expression predictors. The resolution of current arrays is such that candidate genes in affected loci can be rapidly traced and tested further for their involvement if desired or necessary. Reports have appeared showing the prognostic potential of CNVs in several cancer types, including breast,³⁵ gliomas,³⁶ colorectal cancers,³⁷ Wilms’ tumor³⁸ and others (Fig. 5-4). Few studies have been done on radiotherapy patients, although van den Broek and colleagues³⁹ reported finding specific gains and losses in head and neck cancers that correlated with outcome after treatment with combined chemoradiotherapy. It is likely that CGH will complement gene expression and other genome wide assays in defining robust predictors.

FIGURE 5-4 • The prognostic potential of comparative genomic hybridization (CGH). Idbaih et al.³⁶ defined three groups of gliomas based on their patterns of genomic amplifications and deletions that showed prognostic significance for overall survival (A) and progression free survival (B).

(Redrawn from Chin SF, Teschendorff AE, Marioni JC, et al: High-resolution aCGH and expression profiling identifies a novel genomic subtype of ER negative breast cancer, Genome Biol 8:R215, 2007.)

Normal Tissues

Predicting the chance of adverse normal tissue reactions after radiation would also be a valuable aid, allowing dose adjustments and/or protective measures (radioprotective or ameliorating drugs) to be prescribed to patients at risk. It further opens the possibility of increasing tumor radiation doses for the remainder (the majority) of patients, leading to potential increases in cure rates for the patient population as a whole. Several factors are known to affect the response of normal tissues to irradiation. The extent of cell kill of parenchymal cells in the organ at risk is clearly important but is not the only factor. Cytokines are known to play an important role,^40,41 TGF-β being a key cytokine in fibrosis development, influencing fibroblast proliferation and differentiation.

Attempts have been made to predict the chance of adverse reactions from the transcriptional response of patient’s lymphocytes irradiated ex vivo.⁴² The underlying assumption is that the genetic profile of the patient will affect radiation response in all tissues and can therefore be monitored in easily accessible peripheral blood lymphocytes. The further assumption is that sensitivity is best monitored not necessarily by the basal level of gene expression but by transcriptional response to radiation. In the Svensson study,⁴² an expression signature in ex vivo irradiated lymphocytes was identified that discriminated prostate cancer patients with severe late complications following radiotherapy (over-responders) from patients without such complications (nonresponders). Lymphocytes have also been used with some success for predicting radiation induced complications using colony survival⁴³ and apoptosis assays,⁴⁴ supporting the use of these cells for predictive purposes.

Gene polymorphisms have also been studied in relation to adverse radiotherapy induced reactions. Single-nucleotide polymorphisms (SNP) in several genes have been found to correlate with radiation induced fibrosis,⁴⁵ particularly in the TGF-β gene. Lymphocytes represent ideal test material, since they are easy to obtain and SNPs will be the same in all cells (in contrast to gene expression). SNP predictors of adverse reactions are not yet robust, because the studies carried out to date have been small and only a few candidate genes have been investigated. The method appears promising, but evaluation will have to await the results of several ongoing larger studies looking at genome-wide SNPs.

Radiosensitivity

Human tumor cells in tissue culture exhibit wide variation in radiosensitivity despite being irradiated under standard conditions, indicating the presence of inherent genetic factors influencing the radiation response of mammalian cells. It is expected that tumors comprising cells inherently resistant to radiation will be more difficult to cure with radiotherapy than those comprising radiosensitive cells. It is also likely that patients with radiosensitive tumors may be overtreated by “conventional” radiotherapy, undergoing the unnecessary risk of excessive complications to normal tissue, while some radioresistant tumors are undertreated, and would benefit either from a higher dose, an added therapy, or an alternative therapy. The goal of predicting inherent sensitivity is thus to select out tumors at the extremes of the radiosensitivity spectrum for adjusted or alternative therapies, with the aim of improving cure rates of the population as a whole.

Many radiosensitivity genes have been discovered using knockout, knock-down or chemical inhibitor strategies. Many of these genes are involved in DNA repair. However, in recent years it has become clear that the DNA damage response is highly complex, so that predicting the extent of radiation induced killing for any random cell line or tumor has proven to be difficult. One approach has therefore been to use a genome-wide strategy, not dependent on known data or pathways, in which genetic signatures are sought that correlate with intrinsic radiosensitivity in a series of cell lines or tumors. Torres-Roca et al.⁴⁶ described such an approach on selected cell lines from the National Cancer Institute (NCI) panel and found a small set of genes correlating with radiosensitivity. More recently, Amundson and colleagues⁴⁷ extended the studies to the complete NCI panel of cell lines and found a larger set of genes correlating with intrinsic radiosensitivity, with some overlap with the Torres-Roca set. Khodarev and colleagues⁴⁸ used a different approach combining animal tumor and cell line studies, creating a radioresistant and radiosensitive pair of cell lines, to define a radiosensitivity signature. Little has so far been reported on testing radiosensitivity signatures. Pramana et al.²⁵ tested one of these signatures (Torres-Roca) on a group of head and neck cancer patients treated with radiation plus cisplatin, but no significant correlation was found. Recently, this signature has been independently refined, and the updated signature showed a strong trend (P = .06) that patients with tumors predicted to be radiosensitive had a better outcome after chemoradiotherapy (Torres-Roca J, et al., presented at ASTRO, Boston, 2008).

Hypoxia

It has been known for more than half a century that hypoxic cells are up to three times more radioresistant than normoxic cells. In addition, hypoxic cells in tumors are often more slowly proliferating and harder to reach with drugs, because they often are at a distance from blood vessels, making them more chemoresistant. Studies using glass electrodes to measure oxygen tension in human tumors have shown, firstly, how ubiquitous hypoxia is, and secondly, that it is a negative prognostic factor for all the three current major treatment modalities of surgery, radiotherapy, and chemotherapy.^49–54

Cells react to hypoxia by reducing the expression of many genes. However, in contrast, there are a smaller number of genes that are up-regulated. Many of these are dependent on HIF-1α (hypoxia inducible factor), a protein that is stabilized under hypoxic conditions, leading to accumulation in the cell. HIF-1α, a transcription factor, then switches on transcription of a plethora of other genes that have an HRE (HIF responsive element, a specific short DNA sequence) in their promoter region. This allows the cell to adapt to hypoxic stress by, among others, increasing glucose uptake and stimulating angiogenesis. Such up-regulated genes represent a hypoxic signature. Several such signatures have been defined by growing cells under hypoxic conditions and subsequently measuring their expression profiles. Chi et al.²¹ showed that such in vitro–defined signatures have prognostic significance and are more predictive of outcome (overall and relapse-free survival) in breast and ovarian cancer than present clinical parameters. As expected, patients with tumors showing high expression of hypoxia genes, indicating high tumor hypoxia, did worse.

The cell’s response to hypoxia also depends on the degree and duration of hypoxia. More severe hypoxia and longer times under hypoxia are reflected by altered gene expression patterns. Signatures have therefore been described for both acute and chronic hypoxia, and for different degrees of hypoxia.^21,55 These are potentially important distinctions, since there is accumulating evidence that acute hypoxia is more dangerous than chronic hypoxia (equally radioresistant, more viable, more DNA repair–proficient).^56,57 Indeed, Seigneuric and colleagues⁵⁵ showed that in vitro signatures derived from short exposures to hypoxia (acute) predicted outcome in breast cancers whereas those derived from longer exposures (chronic) did not (Fig. 5-5).

FIGURE 5-5 • The prognostic potential of gene expression signatures for hypoxia. Seigneuric et al.⁵⁵ analyzed cell line data to define signatures for genes up-regulated at early times during hypoxic incubation (“early”; acute hypoxia) or after long times (“late long”) chronic hypoxia. Acute signatures (A) had greater potential to distinguish good and poor prognosis in breast cancer patients than chronic signatures (B).

(From Seigneuric R, Starmans MH, Fung G, et al: Impact of supervised gene signatures of early hypoxia on patient survival, Radiother Oncol 83:374-382, 2007.)

An alternative method of deriving hypoxic signatures was described by Winter et al.⁵⁸ They first chose 10 genes known to be HIF-1α (and thus hypoxia)–dependent. They then looked for other genes whose expression correlated with expression of these 10 “seed” genes across a series of human head and neck carcinomas. In this way they defined a hypoxic metagene of 99 genes. This metagene had prognostic value in both head and neck and breast cancer, supporting the notion that hypoxia is an important negative prognostic factor and that gene signatures can be used to monitor it.

Repopulation

During the course of fractionated radiotherapy, tumor cells that have survived (remained clonogenic) up to that point can begin to divide and will increase the number of cells to be killed with the remaining dose fractions. Such tumor cell repopulation is a particular risk during weekends and other gaps in therapy. Cell kill by radiotherapy is thus counteracted by cell production (repopulation) in the treatment gaps. The greater the capacity of a tumor for repopulation, the greater the effective resistance of that tumor will be. Further, the more gaps in treatment there are, the greater the opportunity will be for repopulation. Evidence for this is strongest in head and neck cancer where a number of studies on the influence of treatment time, including split dose and accelerated fractionation, have consistently shown either worse local control if the treatment time is longer, or that higher radiation doses are needed for a given level of local control.^1–4 This is most likely to be due to repopulation, which can therefore limit cure in some cancers. The influence of repopulation will vary between cancers of the same type, and between different types of cancer. Cancers that are in general more slowly growing, (e.g., breast and prostate), may be at less risk from repopulation, although some tumors within these types may be capable of rapid repopulation. It is therefore important to be able to predict repopulation in individual tumors.

Great strides have been made in understanding the molecular events driving and accompanying cell proliferation. Cell culture studies have employed cell populations synchronized in different cell cycle phases, as well as resting cells stimulated to proliferate by serum addition. These have defined a host of cycle phase–specific and proliferation-specific genes. Such proliferation-associated gene signatures include genes that regulate cell cycle progression, such as the cyclin-dependent kinases, and the many external and signal transduction pathways that control entry into and exit from the cell cycle, including cytokines, hormones, growth and antigrowth factors, checkpoints, cell adhesion, and others. Some of these signatures have been shown to have prognostic potential for outcome of breast cancer.^58a In addition, using data-driven approaches, several expression-profiling studies on clinical material have found gene signatures correlating with outcome that contain a preponderance of proliferation-associated genes. Such studies include those on lymphoma,^59,60 breast cancer,⁶¹ hepatocellular carcinoma,⁶² and others.⁶³ These signatures, and those derived from cell culture studies, have thus been shown to have prognostic potential. These have not yet been tested on patients receiving curative radiotherapy as the primary treatment.

Multiple Causes of Resistance

The previous discussion assumes that one gene signature will be sufficient for predicting outcome after a specific treatment such as radiotherapy. This may be too simplistic. Radiotherapy may fail for several reasons: for example, because the tumor is too hypoxia, too radioresistant, or repopulates too rapidly. Furthermore, if it is too intrinsically radioresistant, this may result from increased double-strand break (DSB) repair via homologous recombination, from nonhomologous endjoining, or through down-regulation of apoptosis pathways. Given these multiple reasons for failure, one genetic signature may not be predictive for all tumors. Future developments may therefore include testing each tumor with several signatures, each with proven efficacy and specificity for a biologic process associated with resistance. In this way, the cause of potential resistance in an individual patient can be estimated, leading to an optimum treatment choice.

Combining Assays

No single genome wide assay is likely to be perfect for predictive purposes. For mRNA expression profiling, the most widely used to date, a potential deficiency is that mRNA expression often does not correlate with protein expression, that one would ultimately like to monitor. Nor does it measure posttranslational modifications of proteins, which are central to transmission of signals within the cell, and therefore important for all aspects of cell behavior, including response to damage. Despite this, expression signatures are clearly successful at discriminating cell types, tissue types, tumor subtypes and show promise as response predictors.

At the DNA level, amplifications and deletions detected by CGH do not indicate which genes on these genomic loci are the most important, although the increased resolution of current CGH arrays coupled with the availability of complete DNA sequencing of the genome have facilitated the tracing of relevant genes. A second potential confounding factor is that amplifications do not always correlate with increased gene expression. However, several CGH signatures have been shown to have prognostic potential in a variety of cancers, although all need further validation. Methylation profiles indicate which genes are deregulated in cancer, affecting gene expression. Methylation assays are now becoming genome wide, but whether they will provide additional information to gene expression assays remains to be tested. This may depend on the extent of methylation at a given locus.

Given the potential deficiencies of individual assays, a combination of assays may prove superior. Indeed, combining CGH data with expression profiling allowed Adler and colleagues to separate relevant from irrelevant genetic changes and discover the most important genes driving the “wound” signature, a set of genes with prognostic significance in breast cancer.⁷⁰ Other studies have also shown the added value of combining expression profiling with CGH^35,71 for defining better predictors and better subclassifications of tumors. Combining three approaches (CGH, methylation, and mRNA expression) is now possible, as shown by Sadikovic and colleagues,⁷² providing a more complete picture of genetic and epigenetic changes.

Stem Cells

Another important consideration for predictive assays concerns stem cells. There is good evidence that a small proportion of tumor cells are stem cells and are the only important cells that need to be eradicated for tumor control. In addition, there is evidence that they may be more resistant to treatment by both radiation and drugs than bulk tumor cells. Furthermore, it is also possible, if not likely, that their genetic profiles differ, particularly their gene expression patterns. The relevance of measuring expression in bulk tumor cells can therefore be called into question. This problem is likely to be more important for predicting response where the minority stem cell fraction is crucial than for tumor subtyping, where bulk tumor also defines the subclass. The extent of this problem is as yet unknown, since some prediction success has already been obtained by measuring genetic parameters on bulk tumor. However, it is possible that better predictors can be found by isolating and measuring only cancer stem cells. This awaits better technical procedures and understanding of stem cell characteristics.

Radiosensitivity

The most relevant measure of radiosensitivity is based on the fraction of cells surviving a particular radiation dose, defined as the ability of a cell to undergo at least six doublings, thus forming a clone of at least 50 cells. This is termed the colony-forming, or clonogenic, assay. A summary of predictive assays studies for radiosensitivity is shown in Table 5-1. The most convincing study is that of West and colleagues^73,74 on cervix carcinomas treated by radiotherapy alone. Explanted tumor cells irradiated in vitro had SF₂ values that correlated with outcome. Patients with tumors exhibiting SF₂ values higher than the median value (radioresistant) had significantly worse local control and significantly worse survival rates than did those with tumors with SF₂ values below the median. This trend was the same for all tumor stages. Two of the larger studies on head and neck tumors also showed a positive correlation of in vitro radiosensitivity with local control.^75,76 These clinical studies support the notion that in vitro measurements of radiosensitivity, with all their potential limitations, have relevance to the response of tumors in situ.

Although these results are encouraging, it is unlikely that either the colony assay or similar assays could be used as predictors for routine clinical application, because they take several weeks to complete—unacceptably long for many radiotherapy departments. They also require a highly skilled laboratory team with extensive experience. Other assays have therefore been sought that are more rapid and more suitable for a routine clinical laboratory. These alternative assays are indirect, measuring parameters that have shown correlations with cell kill. DNA DSBs are thought to be the most important and toxic DNA lesion after irradiation. Techniques for their measurement can be completed within a few days rather than the few weeks necessary for colony assays. These include gel electrophoresis and, more recently, antibody detection of a nuclear histone protein that becomes phosphorylated at DSB sites, called γH2AX. The latter provides a method of measuring breaks in tumors irradiated in situ. However, results of studies correlating DSB induction or repair with cell kill have been variable, suggesting that DSBs will not be a reliable predictor of radiation-induced cell kill.

Ionizing radiation also induces chromosome aberrations, including fragments, translocations (dicentrics or reciprocal), rings, chromatid exchanges, gaps, complex types, and micronuclei, all being dose related. Many studies have shown a good correlation between chromosome damage and cell kill.^77–79 Chromosome aberrations can also be measured in a matter of days and can be detected in cells after repair with doses less than 1 Gy, making it as sensitive as γH2AX detection of DSBs. However, although chromosome damage assays have a lot of attractive features as radiosensitivity predictors, ex vivo culturing and irradiations would be required, making it unlikely to prove robust enough for routine clinical use.

Repopulation

Here the goal is to predict, before treatment begins, which tumors are capable of rapid proliferation during treatment. These could be then selected for adjusted radiotherapy schedules or alternative or extra-treatment modalities. Several methods have been tried, including simply counting the frequency of mitoses. Flow cytometry has advantages over counting cells under a microscope, allowing the quantitative measurements of many thousands of cells per minute. By using fluorescent dyes that bind to DNA, DNA histograms can be generated and analyzed for the fraction of cells in each cycle phase (G₁, S, and G₂/M). A more functional assessment of proliferation can be obtained using analogs of thymidine, bromodeoxyuridine (BrdU), and iododeoxyuridine (IdU), which are incorporated into DNA during the S phase. Fluorescent conjugated antibodies allow the degree of analog incorporation per cell to be rapidly measured by flow cytometry. Cell kinetics can be measured in patients with this method by injection or infusion of thymidine analogs at nontoxic tracer doses. The combination of thymidine analogs and flow cytometry allows rapid measurement of the proportion of labeled cells (labeling index [LI]). In addition, by taking samples a few hours after analog administration, one can determine both the LI and the rate of movement through the S phase (T_S).⁸⁰ The ratio of the two (T_S/LI) approximates the potential doubling time, T_pot, a parameter describing the cell number doubling time of a tumor population in the absence of cell loss. Staining and measuring can be accomplished in 1 day. Disadvantages include the necessity of administering a drug (the thymidine analog) and the inability to reliably distinguish malignant from non-malignant cells in a biopsy.

A multicenter study from 11 different centers was carried out for head and neck tumor patients receiving radiotherapy alone given in an overall time of at least 6 weeks, with a total of 476 patients.⁸¹ All patients received the thymidine analog prior to treatment. A univariate analysis showed that only LI was significantly associated with local control (P < .03), higher values correlating with worse outcome. T_pot showed no trend. In a multivariate analysis of local control, LI lost its significance (P < .16). Two potential confounding factors in this study were that each center carried out its own flow cytometry and analysis (rather than a standard reference center), and in none of these analyses was an adequate distinction made between normal and malignant cells. This study suggests that LI but not T_pot may predict repopulation during radiotherapy, but not strongly.

Finding a proliferation marker that does not require administration of a potentially toxic substance remains a worthwhile goal. Such markers include antibodies to Ki67 (cycle-specific), PCNA (S phase–specific),⁸² cyclin A (S/G2 phase–specific)⁸³ and DNA polymerase alpha (cycle-specific).⁸⁴ These all provide static parameters and can be measured by either immunohistochemistry or flow cytometry. The rapidly increasing knowledge of cell cycle control gives hope that expression profiles will be found that can predict repopulation capacity.

Ideally, measurements of proliferation during, and not before, treatment are desired, since this is when the dangerous repopulation takes place. Labeling measurements can be made during treatment, but the data will be strongly dominated by doomed and dying cells that constitute the vast majority of cells after the first few 2 Gy fractions. Such measurements are therefore likely to be misleading, as shown by animal studies.⁸⁵ Until ways can be found to distinguish doomed but intact cells from surviving cells, measurements during treatment will remain unreliable at best, and often come too late to change treatment.

In summary, many of the studies mentioned above have indicated the relevance and importance of predicting tumor proliferation for radiotherapy schedules 6 weeks or longer. Better methods and better knowledge of the biology (e.g., role of cytokines and receptors in irradiated tissue) are now needed. Genome-wide assays (see above) are proving promising for achieving these goals.

Hypoxia

The most direct method to date for measuring tumor hypoxia is the use of glass oxygen electrodes inserted into the tumor. Multiple measurements can then be taken along several tracks, allowing the distribution of oxygen tension to be assessed. The mean or median oxygen tension can be calculated, as well as the fraction of values below a cut-off, usually 5 or 10 mm Hg, giving an estimate of the hypoxic fraction. Several studies have correlated such measurements with outcome after radiotherapy.^49–53 These studies have shown remarkable uniformity in that most, but not all, found that pretreatment oxygen tension was a significant prognostic indicator. These included different tumor sites and all three major treatment modalities, although no sufficiently large series have been published for surgery alone or chemotherapy alone. Hypoxia could affect chemotherapy outcome through lower drug concentrations at hypoxic sites, and the fact that hypoxic cells tend to proliferate slower, reducing the effectiveness of many drugs. Exposure to hypoxia can also lead to selection of apoptosis-resistant cells,⁸⁶ and consequently to malignant progression and an increase in metastatic capacity.^87,88 These may be contributing reasons why hypoxia is also a bad prognostic indicator for surgery. A major disadvantage of electrode methods is its invasive nature and its restriction to accessible tumors.

One of the current most widely used alternative methods to electrodes is the administration of a bioreductive drug, in particular, the nitroimidazoles. Such drugs have been shown to be selectively reduced in and bind to hypoxic cells and can be detected with a labeled drug or by antibodies developed against bound products.^89,90 Two of these nitroimadazoles, pimonidazole and EF5, are approved for human use as hypoxic markers.^90,91 Of interest is that the pimonidazole staining fraction in head and neck tumors does not appear to correlate with electrode oxygen measurements in the same tumors.⁹² Possible reasons include the influence of stroma and necrosis on the polarographic measurements, or that one method may be more influenced than the other by acute (fluctuating) hypoxia. Of further interest is that the only study measuring both pimonidazole staining fraction and oxygen tension with Eppendorf electrodes found that neither parameter correlated with outcome in cervix cancer patients treated with radiotherapy alone.⁹³

Several other methods, both direct and indirect, have also been applied in the clinic for measuring tumor hypoxia.⁵⁴ These include noninvasive assessment of hypoxia using the imaging techniques of PET or SPECT or with MRI, and measuring expression of endogenous markers associated with hypoxia, such as HIF-1α and CA9 (see discussion of genetic assays above). Few studies with sufficient statistical power have yet been carried out to test the predictive potential of these techniques for radiotherapy patients. Finally, it should be noted that none of the methods can distinguish between clonogenic and nonclonogenic cells. Extrapolation from changes in the measured hypoxia parameter occurring during treatment to reoxygenation patterns of the hypoxic, clonogenic cells therefore cannot be made with any degree of certainty.

These data collectively imply that hypoxia can limit cure of cancers by radiotherapy and other modalities in at least three cancer sites. Pretreatment hypoxia measurements with oxygen electrodes have shown the best, although not universal, prognostic significance. Results with exogenous markers (pimonidazole, EF5, and others) and endogenous markers (HIF1α, CA9, and others) have shown mixed results as predictors and do not yet appear to be robust. The ability to predict outcome based on hypoxia measurements is therefore, as yet, suboptimal, and improvements will require better knowledge of which type of hypoxia is important (e.g., acute or chronic) and what each technique is measuring. It remains to be seen whether hypoxia signatures derived from genome wide studies (see above) prove to be more reliable predictors and better indicators of how to treat.

Normal Tissues

Several studies have tested the relationship between the in vitro radiosensitivity of either fibroblasts or lymphocytes and the severity of normal tissue reactions. Geara et al.⁹⁴ and Johansen and colleagues⁹⁵ found a significant correlation between fibroblast radiosensitivity and late reactions. These and other studies indicated that colony survival of fibroblasts after in vitro irradiation may predict for late normal tissue damage, primarily fibrosis. However, two subsequent larger studies could not confirm these results.^96,97 In vitro radiosensitivity of lymphocytes, measured either by colony, cytogenetic or apoptosis assays, have been reported to predict normal tissue morbidity in some studies.^43,44,98

Problems with cell-based assays include their technical difficulty and the long assay times. Clinical confounding factors include an often inaccurate estimate of dose in the target tissue.⁹⁹ While cell based assays have been useful in showing that intrinsic radiosensitivity of somatic cells is probably a contributing cause to differences observed between patients in their reactions to radiotherapy, it is unlikely that they will be routinely useful as predictors for reasons stated above. In addition, intrinsic radiosensitivity is not the only determinant of radiation morbidity. There are a number of biologic factors that influence treatment response. For several tissues (e.g., lung, skin, and intestinal mucosa), the involvement of cytokine-mediated multicellular interactions are implicated, including those mediated by interleukins 2 and 6 (IL-2, IL-6), and interferon alpha (IFN-α).¹⁰⁰ Transforming growth factor beta (TGF-β) clearly also plays an important role in generating and modulating tissue fibrosis in many tissues and organs.^40,41,101 Understanding the mechanisms of normal tissue radiation response other than the conventional radiobiologic paradigm of target cell death will ultimately lead to better prediction.

Analogous to tumors, response of normal tissues to radiation will be determined by multiple factors. With enough knowledge of the relevant genes and pathways, looking at expression or polymorphisms in a far wider range of genes than is now being done may provide a viable approach to predicting normal tissue morbidity (see above). Large microarray studies, analogous to those in tumors, have not yet been reported.