New biomarkers for drug-induced liver injury: are they really better? What do they diagnose?

Abstract

It is proposed in this issue 1 by a distinguished group of academic and pharmaceutical industry investigators that serum biomarkers microRNA-122 (miR-122) and the M65 epitope of keratin-18 (K18:M65) may complement alanine aminotransferase (ALT) and even outperform it as sensitive, specific and predictive biomarkers for drug-induced liver injury (DILI). In an interesting and original study, they have retrospectively analysed the sera from two studies on acetaminophen and antitubercular/immunodeficiency treatments that caused mild DILI in recipients, using these new biomarkers. As better sensitivity, they defined slightly more rapid increase in the biomarker values after exposure to the causative agents, and as better specificity, they cite the lack of response of the two new biomarkers to skeletal muscle injury in subjects taking part in an extreme adventure race. There has been great interest in biomarkers, resulting in a flood of recent publications on various aspects of their applications, shown by more than 620 000 PubMed publications and over 40 000 more per year in the last 2 years. The first mention of ‘biomarker’ in a published title dates back to 1980, when UDP-galactosyl transferase was proposed as a potential biomarker for breast cancer 2 and urinary fibronectin proposed for prostatic cancer 3. Serum ALT activity was first developed for the diagnosis of acute myocardial infarction 4, and only later found to be useful for diagnosing and evaluating liver injury 5. It is generally known to have poor specificity when used alone, despite being thought of by some as a ‘gold standard.’ In cooperation with academic research centres in North Carolina and Sweden, analyses of sera gathered for earlier studies on acetaminophen hepatotoxicity in healthy volunteers 6 and in patients being treated for combined HIV and tuberculosis in Ethiopia 7, have been studied retrospectively, adding to the conventional testing of serum alanine and aspartate aminotransferase (ALT, AST) and alkaline phosphatase (ALP) activities, total bilirubin concentration (TBIL), the new tests for miR-122 and K18:M65. Both were found to show fairly good Pearson correlation coefficients with ALT. When mean values for responders to acetaminophen (those showing ALT rises) were compared to miR-122 and K18:M65 daily measurements, the latter two measures began to increase on Day 4 of acetaminophen 4 g/day, as opposed to Day 5 for ALT, which was taken to show increased sensitivity. The evidence for better specificity was obtained from the before-and-after comparisons of sera from 12 volunteers participating in an extreme 48-hour endurance race, where serum ALT, and even more so AST and creatine kinase activities increased, but K18:M65 and miR-122 did not increase 8. These results are interesting, and the selection of studies was clever, but what have we learnt? It has been known for decades that the aminotransferases are widely distributed intracellular enzymes in tissues, where they link carbohydrate and amino acid metabolism. They are not specific to hepatocytes, although richly concentrated there. When liver cells are damaged or necrotic, the enzymes are released into plasma, where they carry out no function at all, but can be coaxed into showing some activity if provided with substrates and cofactors. They are fairly sensitive indicators of hepatocytic injury, but not of overall liver function. However, when there is sufficient injury to enough liver cells that overall organ capability to function is impaired, as in clearing bilirubin from plasma or synthesis of prothrombin, the combination of ALT and TBIL has been found very useful in the evaluation of liver injury and dysfunction, as we have done using the eDISH (evaluation of Drug-Induced Serious Hepatotoxicity) program 9. This combination is very highly specific for liver dysfunction because only the liver functions to remove bilirubin from plasma, by taking it up from plasma where it is carried by albumin, conjugating it with glucuronide and excreting it into bile. However, just showing that the abnormal test results indicate liver injury is not enough; it must be shown at least probably to be caused by the drug that is suspected, and not by some other cause, of which there are many. The diagnosis of what caused the abnormal test results, at least the probable cause, is the key to recognition of DILI. To date, biomarkers have failed to produce evidence of what caused the liver injury. The best method so far developed or available for differential diagnosis of what caused the liver injury is medical reasoning 10, based on collecting additional data and information that permit consideration of the possible causes and exclusion of all but DI, or finding an alternative cause that can be clinically proved and accepted. The very first element in the Zimmerman dictum 11 that ‘drug-induced hepatocellular jaundice is a serious lesion,’ with considerable mortality, are the words drug-induced. That it should be hepatocellular rather than cholestatic is next, for such injuries are more likely serious and possibly rapidly so, while jaundice is an expression for severity, enough injury to enough hepatocytes that those remaining are not able to carry out liver function very well. Use of the combined ALT & TBIL increase, with clinical information on the time course and other information to make a clinical diagnosis of cause is now the current gold standard for DILI, replacing just ALT alone. We currently think of serious hepatotoxicity causing disability from work, hospital care, secondary renal or brain dysfunction, acute liver failure, death or need for transplantation, and not just an elevation of ALT activities above some locally defined upper range of normal. The diagnosis of DILI is generally not easy and cannot be done reliably just by serum chemistry test elevations nor by liver biopsy. It requires clinical consideration of all the many possible alternative causes, and is a diagnosis of exclusion, after those many causes have been determined to be unlikely or only possible. Unfortunately, there is no quick and easy biomarker that gives the answer accurately and easily. It requires careful evaluation of clinical information often or usually not gathered in routine case reports, and also requires the process of medical differential diagnosis as practised by skilled and experienced physicians who trained since early days in medical school to search for the most likely causes of problems, so they can be appropriately treated. Specialists in the preclinical sciences, brilliant as they may be, are not usually trained in diagnosing. The accurate diagnosis of the probable cause of DILI is made even more difficult because it is usually rare or very rare, especially if serious levels of DILI are considered. Great advances in preclinical drug development have led to marked reduction in drugs being brought to large clinical trials that are likely to cause serious hepatotoxicity in humans, where the problem shifts from which drug is more toxic to which individual patient may be especially susceptible. Large controlled clinical trials are extremely costly and time-consuming, and even as done cannot simulate exactly the numbers and conditions of exposure to approved drugs that may be prescribed in many more and less well-selected or observed patients. Even epidemiological safety nets for detecting reported adverse drug effects after approval, prescription and marketing only capture a small fraction of the total adverse events that occur, and the spotty voluntary reporting usually provides very poor information for making retrospective diagnoses of either causality or severity. It has proved effective, at least so far, to raise the consciousness of sponsors developing new drugs and regulatory staff reviewing them for approval about the advances in understanding idiosyncratic DILI. As we have learnt slowly, even drugs that show an apparently clear dose-related increase in hepatotoxicity can be shown to have a strong and important idiosyncratic factor depending on individual, personal characteristics of recipients. The frequently observed adaptation by liver cells to drugs that initially cause some injury is perhaps an extension of the remarkable ability of the liver to regenerate, quickly regrow to its original size and function and then stop. This is an area for promising new investigation. The challenge for developers and investigators of new biomarkers is to appreciate the difficulty of getting to the correct causal diagnosis, not just to look for biomarkers of some kind of liver injury. The chase after increased sensitivity of biomarkers seems somewhat misplaced, for the rarity of the real problem, serious DILI, demands most extremely high specificity, to avoid ruining the biomarker positive test values with false-positive test results that overwhelm the true positive results. Even for a moderately infrequent serious adverse effect, as for isoniazid where one or two per 1000 patients may show susceptibility to progressive hepatotoxicity, compared with perhaps 15% who showed initial transaminase rises, the importance of very high specificity of biomarker testing is apparent. If a brand new biomarker was found with perfect 100% sensitivity that would detect all patients who show toxicity and miss none, even 99% specificity would not even come close to being good enough. Consider the little table below: It is evident that getting only 9% of biomarker test results correct and >90% wrong, is not a satisfactory outcome for a proposed new biomarker. What biomarker do we have that even comes close to 100% sensitivity and >99% specificity? This is a very high hurdle for biomarker developers to overcome. There may be other important values for a new biomarker development that justifies the enormous effort that is being expended. Better insights into mechanisms may well be extremely useful. As for as making true, accurate diagnoses of causality, nothing at present is better than the well-accepted and proved process of medical differential diagnosis. It may also be observed that the ‘value’ of the test result is not referred to as ‘predictive,’ as has been done for decades following that unfortunate choice of adjectives 12. The test results may or may not be indicative, or even diagnostic, but a single test reflects only what has or is happening, not what will or may happen in future. Use of serial testing over time often helps in predicting what is likely to occur. The search for new and better biomarkers is of great interest, and may produce valuable insights that will help us understand the problems of DILI better. As stated last year 13 about the K18 and high-mobility marker of hepatocyte necrosis from acetaminophen injury, the mechanistic biomarkers are of interest, but we are not ‘there’ yet 14. It is not sufficient just to show that liver injury has occurred. To call it, DILI still requires medical diagnosis of the probable cause to be the drug and not something else.