Genetic variants associated with immune-mediated liver injury from checkpoint inhibitors

Abstract

Background:

The clinical features, liver histology, and genetic variants in 57 patients with moderate to severe immune-mediated liver injury from checkpoint inhibitors (ILICI) are presented.

Methods:

Between 2010 and 2022, 57 high-causality ILICI cases were enrolled in the Drug-Induced Liver Injury Network. HLA and selected candidate gene variants were tested for association with ILICI risk compared to the general population and other DILI controls.

Results:

The 57 high-causality cases were attributed to pembrolizumab (16), ipilimumab (15), ipilimumab and nivolumab (13), and other immune checkpoint inhibitors (13) and occurred at a median of 72 days after the first infusion. Median age was 57.8 years, 66% male, and 89% were non-Hispanic Whites. At DILI onset, 53% had hepatocellular, 35% mixed, and 15% cholestatic, with younger patients more likely to have hepatocellular injury. The incidence of ANA, smooth muscle antibody, and elevated IgG levels was low (17%, 23%, and 0%), but corticosteroids were given to 86%. Microgranulomas and hepatic steatosis were seen in 54% and 46% of the 26 liver biopsies, respectively. The HLA alleles associated with autoimmune hepatitis were not over-represented, but 2 host immune response genes (EDIL3 and SAMA5A) and 3 other genes (GABRP, SMAD3, and SLCO1B1) were associated with ILICI (OR: 2.08–2.4, p<0.01).

Conclusions:

ILICI typically arises within 12 weeks of initiating immunotherapy and is self-limited in most cases. Genetic variants involved in host T-cell regulation and drug disposition were identified, implicating these pathways in the pathogenesis of ILICI. If validated, these findings could lead to improved diagnostic instruments and possible treatments for ILICI.

BACKGROUND

Immune checkpoint inhibitors (ICIs) that block the cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) pathway or inhibit the programmed cell death protein 1 (PD-1) or anti-programmed death ligand 1 receptor have revolutionized modern oncology care and demonstrated superiority over earlier medications in numerous solid organ tumors.^1,2 Currently, 2 CTLA-4 inhibitors, 2 PD-1 inhibitors, and 4 anti-PD L1 inhibitors are approved by the US Food and Drug Administration and are parenterally administered either alone or in combination.³ ICIs are also associated with a multitude of immune-related adverse events (irAEs).⁴ Immune-mediated liver injury from checkpoint inhibitors (ILICI) most commonly arises within 6 months of treatment initiation and is characterized by asymptomatic serum aminotransferase elevations of 1.1–5 times the upper limit of normal (ULN) (ie, grade 1–2).^5,6 However, some patients may develop a cholestatic pattern with a predominance of alkaline phosphatase (Alk P) elevation and less frequently present with jaundice or an elevated international normalized ratio.⁷ Fortunately, the majority of patients with mild to moderate ILICI improve with drug interruption or discontinuation but up to 18% of patients on 2 ICIs develop severe ≥grade 3 toxicity (alanine aminotransferase [ALT] >5 × ULN) toxicity that usually requires high dose corticosteroids for disease control.⁸

The Drug-Induced Liver Injury Network (DILIN) is an ongoing multicenter study of the etiologies and outcomes of DILI in American patients.⁹ With an increasing proportion of DILI cases attributed to ICIs following their approval in 2010, we set out to determine the clinical, genetic, and histological correlates of susceptibility and outcomes in 57 high-causality ILICI cases enrolled in the ongoing DILIN prospective and retrospective registry studies. Since the underlying mechanism(s) of ILICI susceptibility remains unknown, genetic predisposition involving polymorphisms in human leukocyte antigen (HLA) alleles, candidate genes previously associated with idiosyncratic DILI (ie, PTPN22, ERAP-2, and ERAP-1), or other irAEs were explored by comparing ILICI cases to ethnicity-matched controls and to other DILI controls.^10,11

METHODS

DILIN prospective study

Patients with liver injury were enrolled in the ongoing DILIN prospective or retrospective observational registry studies.⁹ Liver injury onset was defined as the first date on which a subject met one of the predefined laboratory criteria of either a serum aspartate aminotransferase or ALT level that exceeded 5 times ULN (or 5 times pretreatment baseline if baseline abnormal), a serum Alk P that exceeded 2 times the ULN (or 2 times pretreatment baseline if baseline abnormal), a total bilirubin of 2.5 mg/dL or above, or an international normalized ratio >1.5 on 2 consecutive blood draws obtained at least 24 hours apart. These criteria equate to a Common Terminology criteria for Adverse Events hepatotoxicity grade of 3 or 4. All research was conducted in accordance with Declarations of Helsinki and Istanbul and approved by local ethics and IRB committees. All study participants provided written informed consent, and prospective patients were enrolled within 6 months of DILI onset.

In the prospective study, a detailed medical history was obtained at the baseline study visit, and additional laboratory and radiological testing were performed to exclude competing etiologies via testing for hepatitis A, B, C, E, HIV, antinuclear antibody and smooth muscle antibody titers, cytomegalovirus, and Epstein-Barr virus infection at the local laboratories. Cases without HEV testing or without HCV RNA results were tested centrally using stored serum samples. Enrolled patients returned for a follow-up study visit 6 months after initial enrollment and those with evidence of chronic injury returned for further study visits up to 48 months.⁹ Chronic DILI was defined as having a persistently elevated serum aspartate aminotransferase, ALT, or Alk P level; histological evidence of liver injury, or clinical evidence of portal hypertension at 6 months after DILI onset.⁹

DILIN retrospective study

Retrospective DILIN registry study participants were required to have a liver injury episode after 1994 and meet the same laboratory criteria as the Prospective Registry, but only a single-study visit was completed. In addition, because subjects may have been enrolled >6 months after DILI onset, complete testing for competing etiologies of liver injury was not required, but all available data, including local diagnostic serologies, liver imaging, and serial liver biochemistries were recorded, and reviewed during case adjudication.

The severity of the DILI episode was categorized on a 5-point scale of mild (1), moderate (2), moderate-hospitalized (3), severe (4), and fatal (5), where a fatal score was assigned only if the patient died or had liver transplant due to DILI within 6 months of onset.⁹

Liver histopathology

Available liver biopsies were reviewed centrally by a single expert liver histopathologist (David Kleiner). All samples were scored for multiple histological features as well as an overall pattern of liver injury.¹²

Causality assessment

The causal relationship between the liver injury episode and the suspect drug was evaluated by the DILIN causality committee. A DILIN expert opinion causality score ranging from 1 (definite >95% likelihood), 2 (highly likely 75%–95% likelihood), 3 (probable 50%–74% likelihood), 4 (possible 25%–49% likelihood) to 5 (Unlikely <25% likelihood) was assigned by consensus agreement of committee members. In subjects with 2 or more implicated drugs, an overall causality score was assigned to the case, and then an individual causality score for each drug was assigned. Only cases with a causality score of 1, 2, or 3 ie, “high-confidence” DILI cases, were included in the analysis. In addition, the Roussel Uclaf causality assessment method score was calculated by the site principal investigator.¹³ A positive rechallenge was defined as any patient who was retreated with the suspect ICI and developed recurrent liver injury with an aspartate aminotransferase or ALT >5 × ULN or abnormal baseline or an Alk p>2 × ULN or abnormal baseline.

HLA and genotype data

HLA sequencing was performed on DNA extracted from the whole blood of patients with DILI using the Illumina MiSeq platform at the Vanderbilt University Medical Center Immunogenomics, Microbial Genetics, and Single Cell Technologies core as described¹⁰ (Supplemental Materials, http://links.lww.com/HC9/B26).

Statistical analysis

Descriptive statistics were computed for demographic and patient characteristics by frequency (percentage) for categorical variables and mean, SD, median, and range for continuous variables. For HLA and candidate gene association analysis, cases were compared to population controls assembled from 5 dbGaP data sets and disease controls from DILI cases due to other drugs, respectively. All class I and II HLA alleles with allele frequency (AF) ≥0.05 in ILICI cases were analyzed. We also assessed if the HLA-DRB1*03:01, HLA-DRB1*04:01, and DRB1*15:01-DQB1*06:02 haplotype previously associated with autoimmune hepatitis (AIH) was associated with ILICI.^14,15 Univariate logistic regression was performed to test age, sex, and principal components 1 to 10 (PC1–PC10, Supplemental S1, http://links.lww.com/HC9/B26). Sex and PC1 met p<0.05 were chosen and adjusted in the multivariable logistic regression models as covariates to test HLA allelic association with ILICI. HLA-DRB1*15:01-HLA-DQB1*06:02 haplotype association with ILICI was performed using haplo.stat. For analysis of 30 candidate genes from the literature associated with other irAEs or DILI (Supplemental Table S1, http://links.lww.com/HC9/B26), variants were excluded if they were (1) multinucleotide variants, (2) monomorphic variants, and (3) with MAF <0.05 in cases. False discovery rate based on Q-value was computed to correct for multiple testing. Significant HLA or single nucleotide polymorphism association was determined based on q-value <0.15. Other HLA alleles or SNPs with P<0.01 were considered suggestive. A few stratification analyses were conducted for the top SNPs identified, comparing (1) chronic to nonchronic DILI cases, (2) melanoma ILICI cases to other tumor types, (3) ILICI cases exposed to statin to those not exposed to statin. Fisher exact tests were performed to compare AF differences between groups.

RESULTS

Incidence of ILICI

Between 2004 and November 2022, 2481 and 248 patients were enrolled into the DILIN prospective and retrospective studies, respectively. An ICI was implicated in 54 prospective cases and 11 retrospective cases. Following case adjudication, there were 57 unique high-causality ICI DILI cases (ie, DILIN causality score of 1, 2, or 3) arising in 56 patients (1 patient had 2 episodes; initially ipilibumab and then ipilibumab/pembrolizumab (study ID # 25 and #26). All 56 patients were enrolled in either the prospective (86%) or retrospective studies (14%). There were also 69 high-causality non-ICI chemotherapeutic agent cases as a comparative group of DILI in oncology patients. The proportion of ILICI cases significantly increased over time from 0% in 2010 to over 14% of all DILIN cases in 2022 (p=0.002) while the proportion of non-ICI chemotherapy DILI cases remained stable (p=0.19) (Figure 1). The 57 high-causality ILICI cases were significantly more likely to be asymptomatic men with hepatocellular or mixed DILI compared to the 69 non-ICI DILI cases (Supplemental Table S2, http://links.lww.com/HC9/B26). The initial and peak total bilirubin levels were also significantly lower in the ILICI cases, whereas the proportion of patients with peripheral eosinophilia was higher (33% vs. 16%, p=0.025), but overall 6-month mortality was high in both groups (42% vs. 54%).

FIGURE 1.

Proportion of cases enrolled in Drug-Induced Liver Injury Network attributed to ICI and non-ICI chemotherapy agents from 2004 to 2022. Abbreviation: ICI, immune checkpoint inhibitor.

High-causality ILICI cases

The implicated agents in the 57 ILICI cases were pembrolizumab,⁴ ipilimumab,¹⁵ ipilimumab combined with nivolumab,¹³ nivolibumab,⁵ durvuluamb,² avelumab,² atezolizumab,² and other combinations.² The median duration of ICI use was 72 days (r: 1–515) with 51% of the high-causality ILICI cases having a hepatocellular injury pattern at onset and the remaining 35% being mixed and 14% cholestatic (Table 1). Median patient age was 57.8 years; 67% were male, and 89% were non-Hispanic Whites. The majority of patients were being treated for advanced-stage melanoma (54%), lung cancer (9%), and various other solid organ tumors (37%). At DILI onset, the median ALT was 399 U/L (r: 53–3219), Alk P of 187 U/L (r: 61–1126), and total bilirubin of 1.0 mg/dL (r: 0.2–21.2). The severity of the liver injury was mild in most patients, with only 20 (35%) having a peak total bilirubin exceeding 2.5 mg/dL. Patients under 40 years old were significantly more likely to present with an acute hepatocellular injury profile compared to older individuals, but their outcomes were otherwise similar (data not shown) (Figure 2). In addition, the 27 patients with ILICI who received ipilimumab-containing regimen were significantly younger (55 vs. 64 y), had a shorter latency to onset (66 vs. 95 d), and were more likely to have a hepatocellular injury pattern at onset (70% vs. 33%) compared to the 30 patients with ILICI attributed to antiPD-1/anti-PDL1 monotherapy (data not shown). A comparison of the 13 patients receiving combination Ipilimumab and nivolumab to those receiving ipilimumab,¹⁵ nivolumumab,⁵ or pembrolizumab⁴ monotherapy showed no difference in drug latency, initial or peak ALT, Alk P or total bilirubin, or overall severity score (data not shown). Finally, the drug latency, initial and peak ALT and Alk P levels, as well as the frequency of chronic DILI and severity scores, were similar in the 14 patients with baseline hepatic metastases compared to the 43 without hepatic metastases (data not shown).

TABLE 1.

Presenting clinical features and outcomes of 57 high-causality ILICI

	Hepatocellular N=30	Mixed N=19	Cholestatic N=8	p
Age (y)	55.5	62.5	71.2	<0.001
Male, n %	19 (63)	13 (68)	6 (75)	0.81
Race, n (%)
Non-Hispanic White	26 (87)	17 (90)	8 (100)	0.85
Non-Hispanic Black	3 (10)	1 (5)	0	—
Asian	0	1 (5)	0	—
Other	1 (3)	0	0	—
Hispanic, n (%)	2 (7)	0	0	0.64
Known liver mets, n (%)	11 (38)	2 (10)	1 (12.5)	0.057
Tumor type, n (%)
Melanoma	20 (69)	8 (40)	3 (37.5)	0.14
Lung cancer	2 (7)	3 (15)	0	—
Other	7 (24)	9 (45)	5 (62.5)	—
Drug latency (d) (range)	68 (1–252)	72 (15–352)	161 (15–515)	0.13
DILI onset labs
AST (U/L)	360	245	142	0.032
ALT (U/L)	620	337	193	<0.001
Alk P (U/L)	133	338	479	<0.001
T. bili (mg/dL)	1.1	0.9	1.3	0.86
INR	1.2	1.1	1.1	0.47
ANA + (%)	4 (14)	3 (20)	2 (25)	0.68
SMA + (%)	4 (14)	3 (20)	5 (62)	0.023
IgG (g/dL) (range)	975 (498–1739)	890 (232–1263)	1238 (651–1810)	0.25
Eosinophils (/uL)	375	260	540	0.38
Eosinophils >500, n (%)	11 (37)	4 (21)	4 (50)	0.29
Corticosteroids, n (%)	27 (90)	17 (90)	5 (62)	0.14
Azathioprine or MMF, n (%)	16 (55)	6 (30)	6 (75)	0.072
Peak labs
AST (U/L)	505	313	219	0.037
ALT (U/L)	1040	597	289	<0.001
Alk P (U/L)	179	573	593	<0.001
T. Bilirubin (mg/dL)	1.7	1.6	1.8	0.89
DILIN severity, n (%)
1 = Mild	22 (73)	12 (63)	5 (63)	0.43
2= Moderate	1 (3)	0	1 (12)	—
3= Moderate- hospitalized	4 (13)	6 (32)	1 (12)	—
4= Severe	2 (7)	0	1 (12)	—
5= Death	1 (3)	1 (5)	0	—
Death, n (%)	9 (35)	8 (50)	2 (43)	0.56
Liver-related death (%)	0	0	0	nc
Chronic DILI (%)	4 (17)	4 (29)	3 (43)	0.32

Note: Data presented as median (range) or n (%).

Abbreviations: Alk P, alkaline phosphatase; ALT, alanine aminotransferase; ANA, antinuclear antibody; AST, aspartate aminotransferase; INR, international normalized ratio; MMF, mycophenolate mofetil; nc, not calculated; SMA, smooth muscle antibody; T. bili, T. bilirubin.

FIGURE 2.

Impact of subject age on immune checkpoint inhibitor hepatotoxicity phenotype. Over 80% of the subjects <40 years of age presented with an acute hepatocellular injury profile, while 70% of those over the age of 60 presented with a mixed or cholestatic lab profile (p=0.002, trend test).

The overall incidence of detectable antinuclear antibody and anti-smooth muscle antibody were low in all 3 groups of patients (Table 1). Forty-nine (86%) patients received corticosteroids, 2 (3.5%) received azathioprine, 25 (44%) mycophenolate, and 5 (9%) received a calcineurin inhibitor. Although 33% had peripheral eosinophilia at DILI onset, they were not more likely to have a rash but they were more likely to report itching (21% vs. 0%, P =0.01). Furthermore, eosinophilia was not associated with detectable autoantibodies, use of corticosteroids, or 6-month outcomes (data not shown). During follow-up, the majority of patients experienced improvement in their liver biochemistries but there was also a high 6-month mortality in all 3 patient groups of nearly 40%. The majority of deaths were attributed to tumor progression, and none of the deaths were attributed to liver injury. The overall incidence of chronic DILI at 6 months after DILI onset was 24%. There were also 8 patients adjudicated as only possible³ or unlikely⁵ ILICI (Supplemental Text S3, http://links.lww.com/HC9/B26, Supplemental Table S3, http://links.lww.com/HC9/B26).

Liver histology

A total of 29 liver biopsies were available for central review from 24 subjects. Three subjects had more than 1 biopsy; 2 biopsies were prior to immunotherapy, and 3 followed the initial biopsy. The first biopsy was performed a mean of 38 days (range: 0–178 d) after injury onset (Table 2) and a variety of histological features were observed. Five demonstrated portal and lobular hepatitis (one with fibrin ring granulomas) with mild to moderate activity, while 2 had predominantly granulomatous hepatitis with fibrin ring granulomas. Five had steatosis or steatohepatitis on biopsy. One other patient (study ID #23) biopsied 168 days after onset showed alcohol-associated cirrhosis on biopsy unrelated to ILICI. Interestingly, 6 cases showed cholestatic or biliary injury as the main feature, with lymphocytic cholangitis or chronic cholestatic changes in 3. Two cases demonstrated only diffuse glycogenosis, possibly related to steroid therapy, and one showed only minimal nonspecific changes consisting of minimal steatosis and inflammation. Zone 3 necrosis with mild steatosis and inflammation was seen in a case attributed to pembrolizumab.

TABLE 2.

Liver histopathological findings in immune-mediated liver injury from checkpoint inhibitors

Patient study ID #	Suspect drug	Onset to biopsy (d)	Granulomas	Steatosis	Description	Outcome
1	Ipilimumab	0	Micro	Minimal	Mild acute hepatitis	Recovered
2	Avelumab	1	Micro	None	Lymphocytic cholangitis	Recovered
3	Ipi/Nivo	1	Fibrin ring	Moderate	Granulomatous hepatitis	Recovered
4a	Ipilimumab	2	—	Mild	Mild steatosis with minimal inflammation	Recovered
5	Pembrolizumab	2	Fibrin ring	Marked	Acute and granulomatous hepatitis	Recovered
6 #	Ipi/Nivo	3	Fibrin ring	Moderate	Granulomatous hepatitis	Chronic DILI
7	Atezolizumab	3	—	Minimal	Glycogenic hepatopathy	Recovered
8	Pembrolizumab	4	—	Marked/SH	Steatohepatitis with marked steatosis, mild ballooning injury and minimal inflammation, no fibrosis	Chronic DILI
9	Pembrolizumab	4	Micro	Mild	Acute cholestatic hepatitis with moderate inflammation and mild cholestasis, some changes of chronic cholestasis	Recovered
10	Ipi/Nivo	6	Micro	Mild/SH	Steatohepatitis with mild steatosis, mild ballooning injury, mild inflammation, and perisinusoidal fibrosis	Chronic DILI
11	Durvalumab	8	—	Minimal	Chronic cholestatic liver injury with mild inflammation and moderate cholestasis	Chronic DILI
12	Pembrolizumab	11	Micro	Mild	Mild steatosis and inflammation	Recovered
13	Ipilimumab	23	Micro	Minimal	Acute cholestatic hepatitis with mild inflammation and moderate cholestasis	Recovered
14	Ipi/Nivo	24	—	Moderate	Moderate steatosis with diffuse glycogenosis	Recovered
15	Ipi/Nivo	25	Micro	None	Acute hepatitis with moderate inflammation	Recovered
16	Ipi/Nivo	29	Micro	Minimal	Minimal steatosis and inflammation	Recovered
17	Nivolumab	35	—	Mild	Mild steatosis and glycogenosis	Chronic DILI
18	Pembrolizumab	48	—	Minimal	Glycogenic hepatopathy	Recovered
19	Pembrolizumab	52	Micro	None	Mild chronic hepatitis with nodular regeneration and perisinusoidal fibrosis	Recovered
20	Pembrolizumab	66	Micro	Mild	Mild zone 3 necrosis with mild steatosis and inflammation	Recovered
21	Ipi/Nivo	104	Micro	Minimal	Minimal steatosis and inflammation	Recovered
22	Avelumab	122	Micro	Minimal	Acute hepatitis with mild cholestasis	Recovered
23	Pembrolizumab	168	—	Marked/SH	Cirrhosis (secondary to alcohol), changes of steatohepatitis and mild cholestasis	Chronic DILI
24	Nivolumab	178	Micro	Mild/SH	Chronic cholestasis and steatohepatitis with duct injury	Chronic DILI

^a 2nd biopsy of 4 on subject #4. 1st biopsy was 31 days before onset. #2nd biopsy of 2 on subject #6. 1st biopsy was 85 days before onset.

There was no discernable difference in the injury patterns attributed to different drugs or drug targets (PD-1 vs. programmed death ligand 1 vs. CTLA-4). Inflammatory, cholestatic, and steatotic patterns were seen in all the groups. However, fibrin ring granulomas were only seen in the 3 PD-1 inhibitor cases. Microgranulomas were commonly found in 16 (66%) cases, as was some degree of steatosis. Plasma cells and eosinophils were uncommon, being increased in only 1 and 3 cases, respectively. The only case with advanced fibrosis was the one with cirrhosis due to alcohol. The presence of steatosis involving more than 5% of parenchyma was associated with BMI >25 kg/m² (p=0.023), suggesting that some, if not all, of the steatosis seen in these cases is likely due to concomitant nondrug etiologies.

Chronic DILI and long-term outcomes

Overall, 11 of the 45 prospectively enrolled patients had evidence of ongoing liver injury 6 months after DILI onset. Compared to the 34 patients with resolving DILI, the 11 chronic DILI patients had higher initial and peak Alk P levels (338 vs. 150 U/L, P=0.057, 566 vs. 226 U/L, p=0.048) and a lower R-value at presentation (2.9 vs. 7.2, p=0.054) (Supplemental Table S4, http://links.lww.com/HC9/B26). However, only the pretreatment baseline ALT in the 11 who developed chronic DILI was higher than the 34 without chronic DILI (42 vs. 20, p=0.028), but Alk P levels were similar. Laboratory data were available at months 12, 24, 36, and 48 in 7, 5, 2, and 2 cases, respectively. Serial ALT and Alk P levels tended to improve over time (Supplemental Figure S1, http://links.lww.com/HC9/B26) but remained elevated at month 24 in 4 of the 5 patients.

Two subjects developed secondary sclerosing cholangitis-like changes on magnetic resonance cholangiopancreatography (study ID #17, #27). Both had mild nonspecific changes on liver biopsy and had been nonresponsive to corticosteroid therapy. Despite the addition of mycophenolate and a calcineurin inhibitor, both experienced radiological progression by magnetic resonance cholangiopancreatography and died of progressive cancer at 4 and 1 years after DILI onset with persistent laboratory abnormalities (Supplemental Text S2, http://links.lww.com/HC9/B26).

Outcomes with rechallenge

Fourteen patients were intentionally rechallenged with the same or different ICI in light of their initial favorable treatment response (Table 3). The median age of these patients was 55 years (range: 27–71); 50% were males, and 64% had melanoma. After a median of 114 days after their initial episode of hepatotoxicity, the 14 patients were rechallenged with a variety of ICI regimens. During follow-up, 6 of the 14 (43%) experienced a positive rechallenge with a median latency to onset that was significantly shorter than the median time to initial DILI onset (15 vs. 74 d, p=0.047). Interestingly, 4 of these 6 patients with a positive rechallenge were receiving maintenance immunosuppression at DILI onset (1 on cyclosporine, 3 on steroids), and 3 required additional immunosuppression. Among the 8 subjects with a negative rechallenge, 3 had a tumor response, 3 died of tumor progression, and 2 were still being treated at the last follow-up.

TABLE 3.

Outcomes with ICI rechallenge in those with prior ILICI

Patient study ID #	Age/gender	Tumor type	Initial ICI (d)	Time to DILI onset (d)	Initial iRAE	Time to 2nd ICI (d)	2nd ICI	Rechallenge outcome
2	60 F	Thymoma	Avelumab (1)	14	Gr 3 liver	98	Avelumab	+ Rechallenge D14 (on CSA)
22	41 F	Thymoma	Avelumab (43)	83	Gr 3 liver	57	Avelumab	+ Rechallenge D55 (on pred); i.v. steroid/ MMF
6	55 M	Melanoma	Ipi/Nivo (29/29)	65	Gr 3 liv	116	Nivo	+ Rechallenge D10 (on pred 60) Pred/MMF Died of tumor progression
29	76 M	Melanoma	Ipi/Pemb (30/99)	112	Gr 3 liv	101	Pem	+ Rechallenge D15 Pred/MMF/CSA (PSC)
30	48 M	Melanoma	Ipi (20)	56	Gr 4	293	Ipi/Pem	+ Rechallenge D20
31	43 F	HCC	Atezo (127)	177	Gr 3	63	Atezo	+ Rechallenge D 10 (tumor progression)
10	71 M	Apical adeno	Ipi/Nivo (42/126)	65	Gr 3 liv/ colitis	67	Nivo	Neg rechallenge (died of tumor progression)
28	28 F	Melanoma	Ipi/Nivo (19/19)	47	Gr 3 liv	241	Pembro	Neg rechallenge
21	49 F	Melanoma	Ipi/Nivo (42/42)	1	Gr 3 liv	95	Nivo	Neg rechallenge (tumor response)
3	27 M	Melanoma	Ipi/Nivo (22/22)	58	Gr 3 Liv	112	Nivo	Neg rechallenge (on steroid)
25	55 M	Melanoma	Ipi/Nivo (63)	76	Gr 3 liver	147	Nivo	Neg rechallenge
16	61 M	Melanoma	Ipi/Nivo (45/45)	63	Gr 3 liver	154	Nivo	Neg rechallenge (Tumor response)
26	55 F	Melanoma	Ipi/Nivo (19/19)	48	Gr 2 liver	177	Nivo	Neg rechallenge (tumor response)
27	61 F	Myoepith	Ipi/Nivo (1/1)	14	Gr 2 liv Pulm	141	Pembro/Ipi	Neg rechallenge (died of tumor progress)
	Med = 55 50% male	64% melanoma	Med = 35.5 d	Med= 60.5 d	—	Med = 114 d	—	—

Abbreviations: Atezo, atezolizumab; CSA, cyclosporine; Ipi, ipilimumab; MMF, mycophenolate mofetil; nivo, nivolumab; Pemb, pembrolizumab; PSC, primary sclerosing cholangitis.

Genetic associations with ILICI

HLA alleles and genome-wide genotype data were available in 51 unique high-causality ILICI cases that included 44 non-Hispanic Whites (NHW), 3 non-Hispanic Blacks, 2 Asians, and 2 Hispanic patients determined by their genetically inferred ancestry. As the majority of cases were NHW, we conducted analysis on all 51 ILICI cases and 44 NHW cases, respectively, compared to untreated population controls (28,130 overall and 17,842 NHW) as well as other non-ICI DILI cases (1459 overall and 1056 NHW). Since univariate analysis showed that sex (p=0.013) and PC1 (p=0.002) met nominal significant association with ILICI, they were adjusted for in multivariable logistic regression models as covariates. For HLA, only HLA-B*13:02 t showed nominal significance (p=0.036 when compared to other DILI cases in NHW). The known AIH alleles in NHW, HLA-DRB1*03:01, HLA-DRB1*04:01, and the overall DR4 (combination of HLA-DRB1*04:01, *04:02, *04:04, *04:07) were also not associated with ILICI (Supplemental Table S5, http://links.lww.com/HC9/B26). Similarly, no significant association was observed for the HLA-DRB1*15:01-DQB1*06:02 haplotype (cases vs. population controls: p=0.42 vs. other DILI cases p=0.63).

The boundary of 25 candidate genes and target SNPs in 5 candidate genes tested are listed in Supplemental Table S1, http://links.lww.com/HC9/B26. Multiple variants with strong linkage disequilibrium in the epidermal growth factor-like repeats and discoidin I-like domains 3 (EDIL3) and SEMA5A genes were significantly associated with ILICI after correction of multiple testing (min q-value=0.13) in the analysis of NHW only (Table 4, Supplemental Figure S2, http://links.lww.com/HC9/B26) and similar results were seen in all 51 ILICI cases. Specifically, the rs1862167 in EDIL3 (OR (95% CI)=2.21 (1.46, 3.34), p=1.83×10⁻⁴) and rs35719165 in SEMA5A (OR=2.08 [1.38, 3.15], p=4.76×10⁻⁴) were the top variants when compared to population controls with consistent results when compared to other DILI cases (Table 4). Three additional suggestive loci (min p<0.01) were GABRP (top variant: rs73800947 p=0.005 and 0.01), SLCO1B1 (rs34234515, p=0.006 and 0.002), and SMAD3 (rs12913535, p=0.009 and 0.037) (Table 4). Finally, the list of 10 genes with SNPs meeting p<0.05 but not considered significant or suggestive loci based on our criteria of p<0.01 are seen in Supplemental Table S6, http://links.lww.com/HC9/B26.

TABLE 4.

Peak variant within the genes that showed genetic association with ILICI in the overall cohort and in the non-Hispanic White group

Overall cohort	Cases (N=51)		vs. population controls (N=28,130)				vs. other DILI cases (N=1459)
GENE	SNP	Allele	Case AF	Control AF	OR (95% CI)	p	Control AF	OR (95% CI)	p
EDIL3	rs1862167	T	0.53	0.38	2.15 (1.39,3.33)	5.73×10−04	0.37	2.13 (1.38,3.27)	5.96×10−04
SEMA5A	rs35719165	A	0.45	0.29	2.26 (1.48,3.46)	1.78×10−04	0.29	2.06 (1.34,3.17)	0.001
GABRP	rs73800947	T	0.16	0.12	2.08 (1.20,3.58)	0.009	0.10	2.01 (1.13,3.60)	0.018
SLCO1B1	rs34234515	A	0.12	0.05	2.41 (1.33,4.39)	0.004	0.05	2.83 (1.38,5.80)	0.045
SMAD3	rs12913535	T	0.16	0.07	1.95 (1.12,3.39)	0.018	0.09	1.86 (1.06,3.30)	0.032
Non-Hispanic Whites	Cases (n=44)		vs. population controls (n=17,842)				vs. other DILI cases (n= 1056)
GENE	SNP	Allele	Case AF	Control AF	OR (95% CI)	p	Control AF	OR (95% CI)	OD P
EDIL3	rs1862167	T	0.60	0.41	2.21 (1.46, 3.34)	1.83×10−04 a	0.40	2.14 (1.43, 3.22)	2.50×10−04
SEMA5A	rs35719165	A	0.45	0.28	2.08 (1.38, 3.15)	4.76×10−04 a	0.29	2.09 (1.36, 3.22)	8.19×10−04
GABRP	rs73800947	T	0.15	0.07	2.32 (1.28, 4.20)	0.0054	0.07	2.38 (1.23, 4.59)	0.0098
SLCO1B1	rs34234515	A	0.13	0.05	2.40 (1.29, 4.49)	0.0059	0.05	3.12 (1.51, 6.44)	0.0022
SMAD3	rs12913535	T	0.18	0.10	2.05 (1.20, 3.53)	0.0092	0.11	1.83 (1.04,3.23)	0.0367

^a Peak variants in EDIL3 and SEMA5A are considered significant after correction for multiple testing with a q-value =0.13 (ie, q-value <0.15).

Abbreviations: AF, allele frequency; SNP, single nucleotide polymorphism.

None of these peak variants nor HLA-B*13:02 distributed differently between chronic and nonchronic ILICI cases (Supplemental Table S7, http://links.lww.com/HC9/B26) or between statin-exposed and nonexposed ILICI cases (Supplemental Table S8, http://links.lww.com/HC9/B26). Although rs34234515 in SLCO1B1 is impacted by concomitant statin exposure, this variant did not reach statistical significance in those with and without statin exposure (p=0.33 and 0.32) (Supplemental Table S8, http://links.lww.com/HC9/B26). Finally, there were 26 ILICI cases with melanoma and 25 with other tumor types. Among the top variants, the rs35719165 in SEMA5A showed significantly lower AF in melanoma patients than others (AF=0.31 vs. 0.6, p=0.005) and remained nominally significant in the NHW group (AF=0.33 vs. 0.6, p=0.02) (Supplemental Table S9, http://links.lww.com/HC9/B26).

DISCUSSION

An increasing proportion of cases enrolled in the ongoing DILIN registry studies are attributed to ICIs, which now account for 15% to 20% of newly enrolled cases as compared to none prior to 2010 (Figure 1). Compared to other chemotherapy agents, ICI cases are less likely to present with or develop jaundice suggestive of a different mechanism of injury, severity, and natural history (Supplemental Table S2, http://links.lww.com/HC9/B26). The suspect drugs implicated in our 57 ICI cases included several different agents, which were administered mainly for metastatic melanoma (54%) and lung cancer (9%). ILICI arose in 82% within 6 months of ICI initiation and within 12 months in 96% of treated patients. The large number of asymptomatic patients (68%) may be due to the routine testing of liver biochemistries prior to each infusion as recommended by practice guidelines.^4,6 Fortunately, only 27% of our patients required hospitalization, and only 33% developed jaundice (>2.5 mg/dL). Prior studies demonstrate that 10%–30% of ICI-treated patients may develop peripheral eosinophilia, which has been previously noted with various iRAEs as well as an improved tumor response.^16,17 However, there was no association between eosinophilia at DILI onset and the presence of autoantibodies, response to corticosteroids or overall outcome (data not shown).

A hepatocellular liver injury pattern was observed in 53% at DILI onset, while 33% had a mixed and 14% had a cholestatic lab profile (Table 1). A significant association between younger patient age and hepatocellular injury was noted (Figure 2) as recently reported with sulfonamide hepatotoxicity.¹⁸ However, the duration of ICI use, liver histology, and response to corticosteroids as well as the likelihood of chronic DILI or death, did not vary by injury pattern at presentation (Table 1). The tendency for older patients to present with a mixed or cholestatic profile is an important clinical observation that may lead to earlier diagnosis and pancreaticobiliary imaging in those patients. The low incidence of detectable antinuclear antibody, anti-SmAb, or elevated IgG levels in our patients suggests a different mechanism of immune-mediated liver injury than that due to AIH. In fact, none of our patients had an IgG level >1500 mg/dL, and only 1 of the 26 patients undergoing liver biopsy had a plasmacytic infiltrate (study ID # 13). Our findings are consistent with prior reports that showed no association between pre-existing autoantibodies and the development of ILICI.¹⁹ Our observation regarding the absence of enrichment of AIH-associated HLA alleles further strengthens the assertion that ILICI is not a variant of AIH but rather a distinct form of indirect DILI.⁶

Although fibrin ring granulomas have been previously reported with anti-CTLA4 hepatitis, the high prevalence of microgranulomas (54%) and hepatic steatosis (44%) has not been previously reported (Table 2).^6,20 Microgranulomas are composed of a cluster of 3 to 6 lobular macrophages that are believed to be involved in the removal of dying or injured hepatocytes. The presence of micogranulomas was not limited to a particular class of ICI, and 1 case related to pembrolizumab showed fibrin ring granulomas, which were previously only associated with anti-CTLA4 therapy.²¹ In prior series, microgranulomas and fibrin ring granulomas from drugs have been associated with a more favorable prognosis. Only 2 of the biopsied cases (study ID #23, #11) had signs of acute liver failure, and neither showed any granulomatous inflammation. We believe that the high incidence of hepatic steatosis likely represents underlying metabolic syndrome or the impact of corticosteroid therapy with associated insulin resistance rather than being a primary manifestation of ILICI.^21,22

During follow-up, most patients had improvement in their liver biochemistries, but 11 (24%) had persistent liver injury at month 6. A large proportion of our patients (40%) also died within 6 months of DILI onset primarily from their underlying malignancy, but none of these deaths were liver or DILI related. Patients who evolved into chronic DILI had a longer drug latency and were more likely to present with a higher Alk P level at onset, which slowly improved during prolonged follow-up (Supplemental Table S4, http://links.lww.com/HC9/B26, Supplemental Figure S1, http://links.lww.com/HC9/B26). Of note, 2 patients developed radiological evidence of secondary sclerosing cholangitis during follow-up that was not apparent at DILI onset²³ (Supplemental Text S2, http://links.lww.com/HC9/B26). This phenotype of liver injury where the inflammation is targeting the intrahepatic and extrahepatic bile ducts has been previously reported with ICIs and other drugs but accounts for only a minority (<1%) of cases.²³ At the time of biopsy, patient study ID #17 did not show evidence of small duct injury or chronic cholestasis.²⁴ Neither of our patients with secondary sclerosing cholangitis responded to immunosuppression, similar to a recent systematic review 26 patients where only 11% responded to corticosteroids.²³ Therefore, endoscopic management of strictures may be the preferred management strategy over intensive immunosuppression once secondary sclerosing cholangitis is identified.²³ Fourteen of our well-characterized subjects were also intentionally rechallenged at a median of 114 days after initial ICI hepatitis with the same ICI agent in most cases (13 of 14) (Table 3). Similar to other published data, 6 patients (42%) had recurrent hepatitis with a substantially shorter latency than the time to the first episode of hepatitis (15 vs. 74 d, p=0.047). The earlier onset of recurrent hepatitis highlights the critical importance of frequent laboratory and clinical assessment following planned rechallenge.^25–27

None of the HLA alleles were associated with ILICI using the overall dataset or by restricting to NHW. In addition, other genes previously associated with all-cause idiosyncratic DILI (PTPN22 and ERAP-1/2) were not associated with ILICI cases.^10,11,28 However, by screening all available SNPs in 30 candidate gene regions previously implicated in published reports of iRAEs, variants in EDIL3 (rs1863267) and SAMA5A (rs35719165) were significantly associated with ILICI after correcting for multiple testing. EDIL3 is a gene of EGF-like repeats and discoidin domain 3, whose protein acts as an integrin ligand.^29,30 The EDIL3 gene loci is associated with various autoimmune diseases and may even be a potential therapeutic target.³¹ The DEL-1/β3 integrin axis promotes regulatory T-cell responses during inflammation resolution, and high levels of EDIL3 are associated with poorer response to ICIs.³² SEMA5A is a member of the semaphorin family, and Nrp-1 (the ligand for the related molecules SEMA3A and SEMA4A) is expressed on and activates regulatory T cells, which are important in preventing ILICI.²⁹ SEMA3A activation also inhibits anti-tumor CD8 T cells. Indeed, one of the hallmarks of exhausted CD8 T cells is coexpression of PD1 and Nrp-1.³³ The previously reported SEMA5A variant (rs3026321) associated with other irAEs in melanoma patients receiving ICI treatment (p=6.31×10⁻⁵) did not show a significant association with ICI hepatotoxicity in our cohort (p=0.394).³³ This may be because 44% of our cohort were nonmelanoma cases or that different variants in SEMA5A may be involved, emphasizing the need for larger studies and replication cohorts.

A few variants in GABRP, SMAD3, and SLCO1B1 were nominally associated with ILICI. GABRP is widely expressed, including in liver tissue, and increased GABRP expression on pancreatic adenocarcinoma cells enhances immunomodulatory and pro-oncogenic infiltration of macrophages.³⁴ SMAD3 expression enhances PD1 on anti-tumor CD8+ T cells and thus reduces anti-tumor immune responses.³⁵ In contrast, SCL01B1 is a human solute carrier organic anion transporter gene that encodes a liver-specific member of the organic anion transporter family.^36,37 Interestingly, SLCO1B1 is involved in the removal of endogenous compounds such as bilirubin and xenobiotics such as statins from blood, and it is highly expressed on the basolateral membranes of hepatocytes. Common and rare variants in this gene are associated with impaired transporter function, statin pharmacokinetics, and statin-related myositis and myalgias.³⁸ The top variant (rs34234515) in SLCO1B1 did not show a different AF between statin-exposed and nonexposed ILICI cases. While the liver is the only tissue in which SLCO1B1 is highly expressed, none of the variants in SLCO1B1 modify its expression as eQTLs in liver tissue.³⁹ Nonetheless, this intriguing discovery of a liver-expressed gene involved in drug transport is clearly biologically plausible and worthy of further study.

Strengths of our study include a large number of high-causality ILICI cases that were enrolled from multiple medical centers. All of the patients underwent extensive testing to exclude competing causes of liver injury and had their cases adjudicated by a panel of DILI experts.⁹ Furthermore, nearly 50% of the patients had available liver biopsy material to review, which further helped to exclude competing causes of liver injury. The absence of an association with AIH-associated HLA alleles further strengthens our impression that ILICI is a distinct phenotype of DILI rather than unmasked idiopathic AIH. Limitations of our study include the lack of a validation cohort to confirm our candidate gene discoveries, although all 5 of our candidate genes are biologically plausible. Therefore, it would be premature to recommend that these variants be used to screen cancer patients for ILICI risk before starting treatment. In addition, drug-treated controls with the same ICIs and tumor types that did not develop ILICI were not available for our genetic analyses. Finally, a multitude of ICI regimens were used in our patients, so additional studies in larger groups of patients receiving the same regimen for the same tumor type are needed since genetic associations in DILI tend to be drug specific. In addition, the lack of whole genome (or exome) sequencing data in our 51 cases precluded us from investigating rare variants.

In summary, ILICI generally presents within 6 months of drug exposure and is responsive to corticosteroid therapy. Nearly one-third of patients develop rapid, recurrent liver injury with ICI rechallenge, highlighting the need for individual patient decision-making regarding the risk versus benefit of rechallenge. Finally, the significant associations with candidate genes involved in T-cell regulation and drug disposition suggest that further studies exploring the role of the adaptive immune response may provide important insights into the pathogenesis of ILICI. However, validation cohorts of a larger number of ILICI cases are needed to confirm our findings, as well as large prospective studies to compare ICI-treated patients who develop ILICI versus those who do not.

Supplementary Material

hc9-8-e0518-s001.docx ^{(685KB, docx)}

AUTHOR CONTRIBUTIONS

All authors contributed to the collection of clinical data, data analysis, and initial and final drafting of the manuscript.

FUNDING INFORMATION

DILIN Grant Acknowledgment: Research reported in this publication was supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) of the National Institutes of Health under award numbers U01DK065211 (Indiana University), U01DK065184 (University of Michigan), U01DK065201 (University of North Carolina-Chapel Hill), U01DK083020 (University of Southern California), U01DK083027 (Thomas Jefferson University/Albert Einstein Medical Center), U01DK100928 (Icahn School of Medicine at Mount Sinai), and U24DK065176 (Duke University). Additional support is provided by the intramural programs of the NIDDK and the National Cancer Institute. The content and opinions expressed in this article are solely the responsibility of the authors and do not necessarily represent the official views of the National Institutes of Health.

CONFLICTS OF INTEREST

Robert J. Fontana receives research support from Kezar and Takeda Pharmaceuticals. Lily Dara receives research support from Intercept. Vincent Chen receives research support from AstraZeneca and KOWA. Raj Vuppalanchi receives institutional funding from Zydus Therapeutics, Eli Lilly, Terns, Astra Zeneca, Takeda, Gilead Sciences, and Galectin Therapeutics. Additionally, he provides consultation services to Fortrea, Medpace, Cour Pharmaceuticals, Regeneron, and GSK, where he serves as a member of hepatic adjudication and data safety monitoring committees.