Peripheral immune factors aiding clinical parameter for better early recurrence prediction of hepatocellular carcinoma after thermal ablation

Abstract

Objectives

Current predictors are largely unsatisfied for early recurrence (ER) of hepatocellular carcinoma (HCC) after thermal ablation. We aimed to explore the prognostic value of peripheral immune factors (PIFs) for better ER prediction of HCC after thermal ablation.

Methods

Patients who received peripheral blood mononuclear cells (PBMCs) tests before thermal ablation were included. Clinical parameters and 18 PIFs were selected to construct Model^Clin, Model^PIFs and the hybrid Model^PIFs-Clin. Model performances were evaluated using area under the curve (AUC), and recurrence-free survival (RFS) were analyzed by Kaplan-Meier analysis and log-rank tests.

Results

244 patients were included and were randomly divided in 3:1 ratio to discovery and validation cohorts. Clinical parameters including tumor size and AFP, and PIFs including neutrophils, platelets, CD3⁺CD16⁺CD56⁺ NKT and CD8⁺CD28^- T lymphocytes were selected. The Model^PIFs-Clin showed increase in predictive performance compared with Model^Clin, with the AUC improved from 0.664 (95%CI:0.588–0.740) to 0.801 (95%CI:0.734–0.867) in discovery cohort (p < 0.0001), and from 0.645 (95%CI:0.510–0.781) to 0.737(95%CI:0.608–0.865) in validation cohort (p = 0.1006). Model^PIFs-Clin enabled ER risk stratification of patients. Patients predicted in Model^PIFs-Clin high-risk subgroup had a poor RFS compared with those predicted as Model^PIFs-Clin low-risk subgroup, with the median RFS was 18.00 month versus 100.78 month in discovery cohort (p < 0.0001); and 24.00 month versus 60.35 month in validation cohort (p = 0.288). Patients in different risk subgroups exhibited distinct peripheral immune contexture.

Conclusions

Peripheral immune cells aiding clinical parameters boosted the prediction ability for ER of HCC after thermal ablation, which be helpful for pre-ablation ER risk stratification.

Keywords:

• Liver cancer
• recurrence
• thermal ablation
• lymphocyte
• peripheral blood

Introduction

Primary liver cancer, including hepatocellular carcinoma (HCC), is the fourth most common cause of cancer-related death worldwide [1]. Image-guided thermal ablation has been considered as a potentially curative treatment for patients, while early recurrence (ER, within 2 years after treatment) and frequent recurrence after thermal ablation are the major causes of patients’ mortality [2,3]. Identifying patients who are at high-risk of ER alerts adequate and timely surveillance, accordingly provide individualized and provident treatment strategies. Classic risk stratification tools mainly focus on some common features, like alpha-fetoprotein (AFP) level [4], multi-nodularity [5] and tumor size [6], which are widely used in clinics while providing limited and imprecise prognosis information. For histological features, such as satellite nodule [7] and microvascular invasion [8], they were helpful for prognosis prediction while post-surgery available and rarely acquired after ablation.

Classic clinical parameters mainly focus on the morphology information of tumor invasiveness, however, the compromised anti-tumor immunity induced by the maladaptive immune responses are the ‘chief culprit’ for HCC progression [9]. Several studies have demonstrated the intratumoral imbalance between the immune-suppressing cells and the immune-enhancing cells could predict prognosis of patients [10,11], on this basis, some immune-subgroups of HCC tumor microenvironment (immune-hot, immune-cold) which composed of different infiltration of immune components demonstrated distinct prognosis of patients [12,13]. Nevertheless, the potential risks of tissue acquisition, exemplifying pain, bleeding, infection and rarely death [14], renders the histopathological immuno-score isn’t easily accessible in routine clinical health care.

Recently, some clinical and preclinical studies have uncovered the circulating immune perturbations that occurred during tumor development [15,16]. Previous studies have indicated the systemic inflammatory markers, like neutrophil to lymphocyte ratio (NLR), platelet to lymphocyte ratio (PLR), C-reactive protein to albumin ratio (CAR), predicted prognosis in HCC and receiving sorafenib therapy [17]. While these markers may not reflect details of immune status of patients. We thus explored the value of readily-available peripheral immune cells of PBMCs and wondered if the peripheral immune factors could reveal prognostic information, and further aid clinical information for better recognition of ER after thermal ablation in HCC patients.

Materials and methods

Patient cohorts and baseline characteristics collection

We retrospectively searched our department’s database from 489 consecutive HCC patients who underwent local ablation and performed the PBMCs tests 3–5 days prior to thermal ablation between 2008 and 2020. The inclusion criteria were as follows: (a) HCC diagnosis confirmed by biopsy or clinical criteria from international guidelines which based on the clinical criteria from the American Association for the Study of Liver Disease (AASLD) and the European Association for the Liver (EASL) guidelines: Clinical noninvasive diagnosis was established by one imaging technique (MR or CT) in nodules above 2 cm showing the HCC radiological characteristic and the two coincidental techniques (MRI + CT) with nodules of 1–2 cm in diameter; (b) all HCC nodules ≤ 5cm in maximum diameter, ≤3 HCC nodules in number; (c) ≤2 HCC nodules if each maximum diameter > 3 (d) pre-ablation routine blood test results were preserved; (e) all patients were guaranteed at least 2-year follow-up; exclusion criteria were as follows: (a) patients with recurrent HCC (b) lost to follow-up within 2 years after thermal ablation; (c) presence of other extrahepatic malignancies or active inflammation; (d) incomplete ablation evaluated by contrast-enhanced ultrasound (US)/magnetic resonance imaging (MRI)/computed tomography (CT) within 1 month after ablation; (e) receiving liver transplantation after ablation. A total of 244 patients were enrolled and randomized in 3:1 to discovery cohort (183 patients) and validation cohort (61 patients) (Figure 1). We collected details on demographic information, tumor characteristics, liver function parameters and PBMCs results obtained 3–5 days prior to thermal ablation from medical records in our institutional database. All the patients were percutaneously treated by a cooled-shaft microwave system (KY-2000, Kangyou Medical, China) or radiofrequency system (WB991029, CelonLab Power, Germany). Written informed consent for thermal ablation procedures and the use of data for research purposes were obtained from each patient before each procedure. Ethical approval was granted by our institutional ethics committee. The approved protocol number in the Human Subjects section is S2017-045.

Figure 1. Study design. PBMCs: peripheral blood mononuclear cells; ER: early recurrence; PIFs: peripheral immune factors; RFS: recurrence-free survival.

Flow cytometry and peripheral immune factors definition

In order to depict the peripheral immune status in more representative ways, we assessed and integrated 18 peripheral immune factors and the baseline levels of each factor were shown in the table (Supplementary Table 1). CD3⁺ was common T cell markers [18], and CD4⁺ as well as CD8⁺ were used to mark T helper cell and cytotoxic T cell [19,20], respectively. In human, naive cells were held to be CD45RA⁺CD45RO^-, and memory T cells to be CD45RA^- CD45RO⁺ [21]. The CD28⁺ was established as a marker of T cell co-stimulation, it was suggested that CD28 engagement by B7-1 could enhance T cell proliferation and interleukin-2 (IL-2) production [22], while CD28^- was be seen as one of markers of immune senescence [23]. Regulatory T cells (Tregs) were marked as CD4⁺CD25⁺ [24]. Natural killer (NK) cells were marked as CD3^-CD16⁺CD56⁺ [25], and natural killer T (NKT) cells were marked as CD3⁺CD16⁺CD56⁺ [26]. Apart from these markers, the total count of white blood cell (WBC) and platelet (PLT), and the percentages of lymphocyte as well as neutrophil were also included. The flow cytometry strategy (Supplementary Figure 4) and antibodies panels (Supplementary Table 2) were described.

Table 1. Baseline characteristics of each cohort.

No. (%)	Discovery cohort n = 183 (75%)	Validation cohort n = 61 (25%)	*p Value
Age(years)			0.933
< 65	134 (73.22)	45 (73.77)
≥65	49 (26.78)	16 (26.23)
Sex			0.482
Female	44 (24.04)	12 (19.67)
Male	139 (75.96)	49 (80.33)
Etiology of cirrhosis			0.704
HBV + HCV	165 (90.16)	56 (91.80)
Other	18 (9.84)	5 (8.20)
Differentiation			0.372
Well	32 (17.49)	13 (21.31)
Moderate	62 (33.88)	15 (24.59)
Poor	14 (7.65)	8 (13.11)
NA	75 (40.98)	25 (40.98)
Tumor number			1.000
Single	135 (73.77)	45 (73.77)
2–3	48 (26.23)	16 (26.23)
Tumor size			0.941
≤2.5cm	89 (48.63)	30 (49.18)
>2.5cm	94 (51.37)	31 (50.82)
BCLC stage			0.904
0	29 (15.85)	10 (16.39)
A	113 (61.75)	39 (63.93)
B	41 (22.40)	12 (19.67)
Child-Pugh			0.751
A	172 (93.99)	58 (95.08)
B	11 (6.01)	3 (4.92)
AFP (ng/mL)			0.452
≤25	115 (62.84)	35 (57.38)
>25	68 (37.16)	26 (42.62)
ALT (U/L)			0.920
≤50U/L	154 (84.15)	51 (83.61)
>50U/L	29 (15.85)	10 (16.39)
AST (U/L)
≤40U/L	148 (80.87)	50 (81.97)	1.000
>40U/L	35 (19.13)	11 (18.03)
WBC(×10^^9/L)			0.881
≤4.8	77 (42.08)	25 (40.98)
>4.8	106 (57.92)	36 (59.02)
PLT (×10^9/L)			0.172
< 100	66 (36.07)	28 (45.90)
≥100	117 (63.93)	33 (54.10)
Lymphcyte (%)			0.880
≤32	74 (40.44)	24 (39.34)
>32	109 (59.56)	37 (60.66)
Neutrophil (%)			0.059
≤60	129 (70.49)	35 (57.38)
>60	54 (26.51)	26 (42.62)
ALB (g/L)			0.522
≤35	27 (14.75)	7 (11.48)
>35	156 (85.25)	54 (88.52)
TBil (umol/L)			0.818
≤17.1	117 (63.93)	38 (62.30)
>17.1	66 (36.07)	23 (37.07)

HBV: hepatitis B virus; HCV: hepatitis C virus; BCLC: Barcelona Clinic Liver Cancer; AFP: alpha fetoprotein; ALT: alanine aminotransferase; AST: aspartate aminotransferase; WBC: white blood cell; PLT: platelet; ALB: albumin; TBil: total bilirubin.

Table 2. Univariate and multivariate analyses of baseline clinical and immune characteristics associated with early recurrence.

Characteristics	Discovery cohort (n = 183)
	Univariate analysis			Multivariate analysis
	HR	95%CI	p	HR	95%CI	p
Age, years (>65 vs. ≤65)	1.135	(0.587, 2.194)	0.707
Sex (male vs. female)	1.409	(0.700, 2.836)	0.337
Tumor number (multiple vs. single)	1.338	(0.690, 2.595)	0.389
Tumor size, cm (>2.5 vs.≤2.5)	2.718	(1.467, 5.037)	0.001**	4.023	(1.944, 8.327)	0.000***
BCLC stage (B–A vs. 0)	1.804	(0.772, 4.214)	0.173
Child-Pugh (B vs. A)	0.757	(0.214, 2.681)	0.666
AFP, ng/mL (>25 vs.≤25)	2.387	(1.296, 4.396)	0.005**	2.279	(1.102, 4.716)	0.026*
ALT, U/L (>50 vs.≤50)	0.794	(0.351, 1.793)	0.578
AST, U/L (>40 vs.≤40)	1.168	(0.557, 2.450)	0.681
ALB, g/L (≤35 vs.>35)	0.917	(0.403, 2.087)	0.836
TBil, umol/L (>17.1 vs.≤17.1)	0.896	(0.486, 1.652)	0.725
WBC,×10^^9/L (>4.8 vs.≤4.8)	1.078	(0.595, 1.953)	0.804
Lymphocyte,% (≤0.32 vs.>0.32)	2.198	(1.202, 4.019)	0.011*
Neutrophil,% (>0.60 vs.≤0.60)	2.933	(1.522, 5.653)	0.001**	3.374	(1.509, 7.546)	0.003**
PLT,×10^^9/L (≤100 vs.>100)	2.365	(1.276, 4.383)	0.006**	2.561	(1.225, 5.354)	0.012*
CD3^+ % (≤69 vs.>69)	1.019	(0.567, 1.831)	0.949
CD4^+ % (≤44 vs.>44)	1.063	(0.588, 1.922)	0.839
CD8^+ % (≤21 vs.>21)	0.980	(0.544, 1.763)	0.945
CD4^+/CD8^+ (>2.2 vs.≤2.2)	0.807	(0.441, 1.478)	0.488
NK % (>7 vs.≤7)	1.814	(0.936, 3.515)	0.078
NKT % (>10 vs.≤10)	5.314	(2.009, 14.056)	0.001**	5.137	(1.657, 15.929)	0.005**
CD19^+ % (≤15 vs.>15)	1.033	(0.575, 1.856)	0.914
Treg % (>7.5 vs.≤7.5)	1.301	(0.721, 2.349)	0.382
CD4^+CD45RA % (>12.5 vs.≤12.5)	0.991	(0.549, 1.789)	0.976
CD4^+CD45RO % (≤30 vs.>30)	0.933	(0.519, 1.679)	0.817
CD4^+CD45RA/CD4^+CD45RO (>0.45 vs.≤0.45)	0.937	(0.340, 2.581)	0.899
CD8^+CD28^+ % (≤1.1 vs.>1.1)	1.339	(0.733, 2.447)	0.342
CD8^+CD28^- % (>12 vs.≤12)	3.067	(1.648, 5.707)	0.000***	3.246	(1.574, 6.695)	0.001**
CD8^+CD28^+/CD8^+CD28^- (≤0.5 vs.>0.5)	1.458	(0.712, 2.986)	0.303

*p < 0.05; **p < 0.01; ***p < 0.001. BCLC: Barcelona Clinic Liver Cancer; AFP: alpha fetoprotein; ALT: alanine aminotransferase; AST: aspartate aminotransferase; ALB: albumin; TBil: total bilirubin; WBC: white blood cell; PLT: platelet.

Follow-up protocol

Within one month after thermal ablation, all patients underwent contrast-enhanced US or contrast-enhanced MRI/CT to guarantee the completely destroy of the tumor vitality zone. During the follow-up period, liver function tests, serum AFP examinations, abdominal ultrasound, contrast-enhanced US/MRI/CT imaging scans were performed once every 3–6 months or earlier if clinically indicated. In cases with suspected distant metastasis, chest CT, whole-body bone scans, or positron emission tomography (PET-CT) were performed selectively. The primary endpoint of the study was the ER which was defined as the interval between the date of thermal ablation and the first date of tumor recurrence on imaging within 2 years.

Statistical analysis

We used the χ² test for categorical data and the two-sample t test for continuous variables. Model construction was used data from discovery cohort and confirmed in validation cohort. The cutoffs of clinical variables for outcome referred as previous articles, and the functional risk form as well as the best cutoffs of immune subsets were investigated with restrictive cubic splines. In discovery cohort, clinical and immune variables were initially selected by univariate analysis and multivariable logistic regression analysis (p < 0.05) to construct predictive models. The variance inflation factors (VIFs) were calculated to test for collinearity, and variables with a VIF greater than 10 were dropped.

Model performances were assessed according to some metrics including area under the receiver operating characteristic curve (AUC), sensitivity (Se), specificity (Sp), positive predictive value (PPV) and negative predictive value (NPV). Differences between models were assessed using DeLong’s test in discovery and validation cohorts. The optimal cutoff value for risk subgroups’ division was determined by Youden’s index in the discovery cohort. Kaplan–Meier analysis and log-rank tests were applied to assess RFS among different risk subgroups. Statistical tests were performed with R software (version 4.1.1) and SPSS Statistics 25 (IBM) and GraphPad Prism 9.0 (GraphPad Software).

Results

Patient characteristics

In discovery cohort, 75.96% of the patients were men and 73.22% of patients were under 65 years old. Besides, 73.77% of HCCs were unifocal, and 77.60% were in BCLC 0–A stage. In validation cohort, 80.32% of patients were in BCLC 0–A stage. Patients in these two cohorts with balanced demographic and clinical characteristics, which were summarized in (Table 1). Baseline levels of PIFs in discovery and validation cohorts were shown in the table (Supplementary Table 1).

This study was censored on October 31, 2022 to guarantee all patients having at least two-year follow-up. In discovery cohort, 42.62% of patients recurred within 2 years, among them 6.41% of the patients developed local tumor progression, and 3.85% of patients developed local tumor progression and intrahepatic recurrence, and 83.33% of patients developed only intrahepatic recurrence. 6.41% of patients developed intrahepatic as well as extrahepatic recurrence. In validation cohort, 49.18% of patients recurred within 2 years, among them 6.67% of patients occurred local tumor progression, and 3.33% of patients occurred local tumor progression and intrahepatic recurrence. 86.67% of patients occurred only intrahepatic recurrence and 3.33% of patients had intrahepatic as well as extrahepatic recurrence.

Construction of Model^Clin and the performance for ER risk stratification

Firstly, we evaluated the predictive value of clinical characteristics. Univariate analyses revealed AFP (>25 vs. ≤ 25 ng/mL) and tumor size (>2.5 vs. ≤2.5 cm) exhibited statistically association with ER outcome in discovery cohort (Table 2). We evaluated the two factors in multivariate analyses (Table 2) and incorporated them into Model^Clin for ER outcome prediction. The formula of Model^Clin was shown in the supplementary (Supplementary Table 4). Results showed that the AUC of Model^Clin were 0.664 (95%CI:0.588–0.740) and 0.645 (95%CI:0.510–0.781) in discovery and validation cohort, respectively (Figure 2(a,b)).

Figure 2. Performances of Model^Clin and Model^PIFs-Clin. (a–b) The AUC and 95%CI of Model^Clin and Model^PIFs-Clin for ER prediction in two cohorts. AUC: overall area under the receiver operating characteristic curve; CI: confidence interval; PIFs: peripheral immune factors; ER: early recurrence.

Construction of Model^PIFs and the performance for ER risk stratification

Next, we explored the predicting values of peripheral immune profiles for the ER outcome of HCC. Firstly, all 18 immune features were evaluated by the restrictive cubic spline function in which the ongoing trends of immune features and the cutoffs to the ER outcome were exhibited (Supplementary Figure 1). Univariate analyses indicated the lymphocytes ≤0.32 vs.>0.32 (p = 0.011), neutrophils > 0.60 vs.≤0.60 (p = 0.001), PLT ≤100 vs.>100 (p = 0.006), NKT > 10 vs.≤10 (p = 0.001) and CD8⁺CD28^->12 vs.≤12 (p < 0.0001) were significantly associated with OS (p < 0.05) (Table 2). Results indicated four variables with multivariable p < 0.05 including neutrophils > 0.60 vs.≤0.60 (p = 0.008), PLT ≤100 vs.>100 (p = 0.011), NKT > 10 vs.≤10 (p = 0.008) and CD8⁺CD28^->12 vs.≤12 (p = 0.002) (Table 2), and the four variables were constructed as Model^PIFs. The formula of Model^PIFs was expressed (Supplementary Table 4). The AUCs of Model^PIFs were 0.737 (95%CI:0.666–0.809) and 0.700 (95%CI:0.589–0.831) in discovery and validation cohorts, respectively (Supplementary Figure 2).

Peripheral immune features together with traditional clinical parameters for hybrid Model^PIFs-Clin construction

The above results indicated either the clinical parameters or the peripheral immune profiles were in association with ER of HCC, therefore we explored whether incorporating both of them enabled satisfied predicting performance. After the collinearity test, all 6 variables were put in multivariable logistic regression to construct the final model (Supplementary Table 3), and all 6 parameters were included with multivariable p < 0.05 (Table 2). The formula of Model^PIFs-Clin was shown: Logit(ER)=1.21619*(Neutrophil=1)+0.94055*(PLT=1)+1.63653*(NKT = 1)+1.17747*(CD8⁺CD28^-=1)-1.39204*(Size=1)+0.82389(AFP=1)-2.85985.

(PS: Neutrophil > 0.60 as 1,≤0.60 as 0; PLT ≤100 as 1,>100 as 0; NKT > 10 as 1,≤10 as 0; CD8⁺CD28^->12 as 1,≤12 as 0; Size > 2.5 as 1,≤2.5 as 0; AFP > 25 as 1,≤25 as 0).

The Model^PIFs-Clin showed an increase in discriminative performance compared with Model^Clin, with the AUC improved from 0.664(95%CI:0.588–0.740) to 0.801(95%CI:0.734–0.867) in discovery cohort (p < 0.0001), representing a 13.70% improvement in AUC over the traditional clinical model; and from 0.645(95%CI:0.510–0.781) to 0.737(95%CI:0.608–0.865) in validation cohort (p = 0.1006), representing a 9.20% improvement in AUC over the traditional clinical model (Figure 2(a,b)). The Model^PIFs-Clin was built as a weighted sum of immune and clinical prognostic variables observed of each patient, which allowed us to derive a patient-specific risk score. For interpretability, we divided all patients into two subgroups according to the optimal cutoff value −0.3783 decided by Youden’s index of discovery dataset. Patients with risk score above −0.3783 were in Model^PIFs-Clin high-risk subgroup and patients below the cutoff value were in Model^PIFs-Clin low-risk subgroup. ER in Model^PIFs-Clin high-risk subgroup was significantly higher compared with Model^PIFs-Clin low-risk subgroup, with the incidence of 68.67% and 21.00% (p < 0.001) respectively, 3.3-times increased in discovery cohort (Figure 3(a)); and 54.29%, 42.31% (p = 0.355) respectively, 1.3-times increased in validation cohort (Figure 3(b)). Kaplan–Meier analysis showed that patients predicted as Model^PIFs-Clin high-risk subgroup had a poor recurrence-free survival (RFS) compared with those predicted as Model^PIFs-Clin low-risk subgroup, with the median RFS was 18.00 month (95%CI:12.93–23.07) versus 100.78 month (95%CI:91.57–109.97) in discovery cohort (p < 0.0001) (Figure 3(c)); and 24.00 month (95%CI:13.74–34.26) versus 60.35 month (95%CI:44.29–76.40) in validation cohort (p = 0.288) (Figure 3(d)).

Figure 3. Model^PIFs-Clin risk categories in discovery and validation cohorts. The cumulative incidence of ER in Model^PIFs-Clin-high subgroup compared with Model^PIFs-Clin-low, with 78.0 and 32.2% (p < 0.001) in discovery cohort (a), and and with 54.29 versus 42.31% (p = 0.355) in validation cohort (b). The recurrence free survival (RFS) in Model^PIFs-Clin different subgroups of discovery cohort (c) and validation cohort (d). ns: not significant; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Model^PIFs-Clin exhibited improved discriminative capabilities

Identifying the at-risk ER patients are particularly significant for their prognosis and post-ablation treatment decision-making, therefore we further evaluated the exact benefit of Model^PIFs-Clin for patients. The Se of Model^PIFs-Clin was 0.731 in discovery cohort, outperformed the Se of 0.603 in Model^Clin with approached statistical significance (p = 0.089). In validation cohort, the improvement was confirmed with the Se of 0.833 in Model^PIFs-Clin compared with 0.533 in Model^Clin (p = 0.012) (Table 3). Model^PIFs-Clin yielded the PPV of 0.687 in discovery cohort compared with 0.553 of Model^Clin (p = 0.074); and in validation cohort, Model^PIIs-Clin yielded the PPV of 0.625, which was comparable with 0.667 of Model^Clin (p = 0.737). Details of model performances were shown in table (Table 3). Additionally, of the ER patients who were underestimated as low-risk patients by Model^Clin, Model^PIFs-Clin could upstage 48.39% (15/31) and 43.75% (7/16) of them in discovery and validation cohorts, respectively (Figure 4).

Figure 4. Model^PIFs-Clin reclassified the ER patients who were underestimated by Model^Clin as low-risk. (a) Among those ER patients who were underestimated as low-risk by Model^Clin, Model^PIFs-Clin could screened out 48.39% (15/31) as high-risk ER patients in discovery cohort. (b) In validation cohort, Model^PIFs-Clin screened out 43.75(7/16) the ER patients who were classified as low-risk by Model^Clin.

Table 3. Prognostic performance of the models.

Model	Discovery cohort			Validation cohort
	Model^Clin	Model^PIFs	Model^PIFs-Clin	Model^Clin	Model^PIFs	Model^PIFs-Clin
AUC	0.664 (0.588–0.740)	0.737 (0.666–0.809)	0.801 (0.734–0.867)	0.645 (0.509–0.781)	0.700 (0.569–0.831)	0.737 (0.608–0.865)
p Value	–	0.192^#	<0.001*	–	0.524^#	0.100*
Sensitivity	0.603	0.603	0.731	0.533	0.633	0.833
Specificity	0.638	0.762	0.752	0.742	0.645	0.516
PPV	0.553	0.653	0.687	0.667	0.633	0.625
NPV	0.684	0.721	0.790	0.622	0.645	0.762

^#Compared the AUC between Model^Clin and Model^PIFs.

*Compared the AUC between Model^Clin and Model^PIFs-Clin.

Model^PIFs-Clin subgroups reflect different peripheral immune profiles

As Model^PIFs-Clin might reflect the distinct peripheral immune contexture that affected HCC recurrence, we analyzed the differences of peripheral immune features in Model^PIFs-Clin subgroups. In order to increase the number of patients for exploratory immune features’ analyses, we pooled the discovery and validation cohorts.

Although the CD8⁺(p < 0.0001) was significantly higher in Model^PIFs-Clin high-risk subgroup compared with the low-risk subgroup (Figure 5(g)), the CD8⁺CD28^- (p < 0.0001) was significantly higher in Model^PIFs-Clin high-risk subgroup compared with the low-risk subgroup (Figure 5(h)) and the CD8⁺CD28⁺ were comparable in two subgroups (Supplementary Figure 3). Therefore, we further compared the ratio of CD8⁺CD28^-/CD8⁺ and CD8⁺CD28⁺/CD8⁺. Results indicated that CD8⁺CD28^-/CD8⁺ were significantly higher in Model^PIFs-Clin high-risk subgroup (p = 0.0035) (Figure 5(i)), while the CD8⁺CD28⁺/CD8⁺ were significantly lower in Model^PIFs-Clin high-risk subgroup (p < 0.0001) (Figure 5(e)). Consistently, the CD8⁺CD28⁺/CD8⁺CD28^- (p < 0.0001) was lower in Model^PIFs-Clin-high subgroup (Figure 5(d)). All these results may suggest the T cell inactivation status in the high-risk subgroup.

Figure 5. Differences of peripheral immune factors in Model^PIFs-Clin subgroups. According to the optimal cutoff by Youden’s Index, we divided all patients into two subgroups, for whose predictive value of Model^PIFs-Clin were below the cutoff were classified in Model^PIFs-Clin-low subgroup, and those above it were classified in Model^PIFs-Clin-high subgroup. The immune contexture was distinct between the two subgroups. ns: not significant; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Inflammation-related feature of neutrophil was significantly higher (p < 0.0001) in Model^PIFs-Clin high-risk subgroup (Figure 5(f)). Conversely, the lymphocyte (p = 0.0024), CD4⁺/CD8⁺ (p = 0.0002) and CD19⁺ (p = 0.0020) exhibited reduced expression in Model^PIFs-Clin high-risk subgroup (Figure 5(a–c)). Specially, the NKT (p < 0.0001) was also significantly higher in Model^PIFs-Clin high-risk subgroup (Figure 5(j)). Other immune features in the two subgroups were shown in figure (Supplementary Figure 3).

Discussion

To date, there is no widely accepted clinical-using predictors for ER of HCC after thermal ablation [27], here, we constructed a hybrid model which combined peripheral immune factors and routine clinical parameters, reflecting both tumor’s extrinsic macro-aggressiveness and intrinsic micro-maladjustment, presented improved discriminative abilities for ER prediction in both discovery and validation cohort. Specifically, the Model^PIFs-Clin reached an AUC of 0.801 in discovery cohort, representing a 13.70% improvement in AUC over the traditional clinical model; and the AUC of combined model was 0.737 in validation cohort, representing a 9.20% improvement. Besides, the hybrid model exhibited enhanced risk discrimination and recognition performances compared with clinical model. Model^PIFs-Clin better screened out up to 48.39% ER patients who were classified as low-risk patients by Model^Clin. Patients in Model^PIFs-Clin high-risk subgroup were 3.3-times increased in ER rate compared with Model^PIFs-Clin low-risk subgroup. These results were supported by validation cohort.

Image-guided thermal ablation technique has evolved considerably in the past 20 years and been established as a reliable treatment option for HCC patients [28]. However, the post-therapy recurrence of HCC, especially for the ER which is defined as tumor recurred within 2 years after therapy, much hampered patients’ long-term benefit. For HCC which is conforming to the Milan Criteria, the ER rate was up to 38.9% after thermal ablation [29]. Valid and easily-applicable prognostic criteria that define subgroups of patients could contribute to better curative strategies in clinical application [30].

Previous studies mainly focused on the in situ immune features of tissue sample, while the intratumoural heterogeneity [9] and the potential risk of the invasive acquisition procedure [14] restricted its widely-using in routine health care. Histopathological predictors of microvascular invasion [8] and satellite nodules [7] which were post-surgery available and rarely obtained for patients treated with thermal ablation. The clinical utility of a risk model depends on its performance and the ease of implementation in the clinic. In contrast to the basic indicators for patients’ prognosis, like tumor size and AFP, which the discriminatory performances are often reported below an AUC of 0.70 [31,32], the immune-clinical predictive model requires only immune features from pre-ablation peripheral blood and common clinical information, but exhibited enhanced discriminatory capability and clinical usefulness. Therefore, the Model^PIFs-Clin not only puts an immune-clinical prognostic paradigm for but also pinpoints the potential of circulating immune contexture to predict disease and therapeutic outcomes of HCC.

Local immune responses in the tumor microenvironment have always been the interest of oncology and immunology fields, yet clinical and preclinical advancements are beginning to focus on systemic perturbations that occur during tumor development and the critical contribution of circulating immune cells to antitumor immune response [15,33]. Previous studies have revealed the systemic inflammatory markers acquired via routine blood tests and blood biochemistry, like NLR, PLR and CAR predict prognosis in HCC and receiving sorafenib therapy [17]. While these limited markers may fail to reflect details of immune status of patients. Researches have indicated that the inactivation of some lymphatic lineages [34,35] and the increase in immunosuppressive myeloid populations [36,37] contribute to the perturbations in tumor development and progression. Our data suggest that patients in Model^PIFs-Clin-high subgroup exhibited higher CD8⁺CD28^- cells and the CD8⁺CD28⁺ cells were comparable in two subgroups, while the CD8⁺CD28⁺/CD8⁺CD28^- were lower compared with low-risk subgroup. The CD28⁺ was established as a marker of T cell co-stimulation and CD28 engagement by B7-1 could enhance T cell proliferation and interleukin-2 (IL-2) production [22], therefore higher proportion of CD8⁺CD28⁺ and ratio of CD8⁺CD28⁺/CD8⁺CD28^- was indicated more potent immune cytotoxicity. While distinguished from CD28⁺, the CD8⁺CD28^- T cells are defined as senescent T cells [38], which elevated some immunosuppressive receptors to play immunoregulatory roles, and the potential pro-tumoral effect of CD8⁺CD28^- cells can be observed in glioblastoma [39] and non-small-cell lung cancer [40]. Thus, the higher proportion of CD8⁺CD28^- may suggest the inactivation status of T cells in Model^PIFs-Clin-high subgroup. Besides, the NKT cells were expended in Model^PIFs-Clin-high subgroup. Generally, NKT and NK cells share similarities in phenotype and function participating in antitumor responses [41,42]. While some studies have indicated that the above-median percentage of NKT-like cells was associated with poor disease-free survival in colorectal cancer, for the reason that reduced expression levels of the natural cytotoxicity receptors NKp44 and NKp46 on the expanded NKT-like and NK cells [43]. Additionally, myeloid-derived suppressor cells (MDSCs), like neutrophils, modulated tumor microenvironment via multiple direct and indirect manners [44], and high neutrophil counts were independently associated with poor overall survival and more-advanced disease [45]. In this study, the levels of neutrophils in Model^PIFs-Clin-high subgroup were higher than the low-risk subgroup, which indicated the neutrophils may also affect patients’ prognosis after thermal ablation via immune-suppressive effects. Other immune features, like CD4⁺/CD8⁺ and CD19⁺, also exhibited differences between the two subgroups. In our study, low-risk subgroup also showed higher ratio of CD4⁺/CD8⁺ than high-risk group, which was consistent with the results of regressing melanoma [46] and responders with metastatic melanoma receiving chemoimmunotherapy [47]. For CD19⁺ B cells’ alternations, researchers have reported that preoperative level of CD19⁺IL-10⁺ Breg cells in HCC patients was significantly lower than healthy donors for the tumor microenvironment may produce abundant chemo-attractants which are responsible for the ‘homecoming’ signals to orient regulatory lymphocytes into the tumor [48]. In our study, the baseline levels of CD19⁺ lymphocytes were lower in high-risk subgroup than the low-risk group, other possible correlations and interventions between B cells and other lymphocytes are worth exploring. Therefore, the localized tumor perturbations cannot exist without continuous communication with the periphery, which suggests the potential therapeutic targets for reversing and reshaping the impaired circulating immune responses.

Our studies have some limitations: firstly, as the matter of statistical power as well as the limited sample size in validation and discovery cohorts, the hybrid model exhibited slightly moderate improvement in validation cohort, which was not as prominent as in the discovery cohort. Secondly, for the reason that the PBMCs tests were performed as early as year of 2008, which may inevitably be influenced by the technology improvement. While the improving trends in the AUC of the Model^PIFs and hybrid model was encouraging. Thirdly, all immuno-phenotyping we applied were classic immune cells, more refined immune cell phenotypes were needed. Besides, we didn’t show the dynamic changes of the peripheral immune cells in different treatment time-points, which may be more helpful in reflecting the dynamic evolution of peripheral immune contexture. At last, for the reason that PBMCs test has not been established in mostly ablation treatment centers, we failed to acquire data from outer validation cohorts, but to some extent, the present results confirmed the promising predictive capabilities in internal validation cohort, further prospective studies were needed to validated the results.

In conclusion, we developed a hybrid model reflecting both immune features and clinical parameters which allowed better prediction of ER status of HCC patients. Given the serum PBMCs analyses are easy to perform, utilizing the PIFs to predict the clinical outcomes of patients with HCC is promising. Going beyond the prognosis prediction, it may be compelling to investigate the Model^PIFs-Clin for predicting and monitoring response to immunotherapy of HCC in the future.

Acknowledgements

The authors thank the site investigators, study coordinators and patients who participated in the trial.

Disclosure statement

The authors declared that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Additional information

Funding

This work was supported by National Scientific Foundation Committee of China under Grants 81971625, Grants 82030047, Grants 92159305, Grants 82227804 and Grants 82102043.