Hypoxia and cribriform growth in prostate cancer – establishing a link via MRI

Abstract

Background

Prostate Cancer (PCa) is a heterogeneous disease, where hypoxia and cribriform growth have been established as adverse prognostic features. Hypoxia refers to a condition of limited oxygen concentration within the tumor microenvironment, associated with enhanced tumor aggressiveness, resistance to therapy, and poor clinical outcome. Likewise, cribriform growth has also been related to poor prognosis, though the biologic basis remains unclear. Recent pathological studies suggest that there is an association between cribriform growth and tumor hypoxia, but evidence from PCa patient studies are scarce. To fill this knowledge gap, this study aims to investigate the association between hypoxia measured by MR imaging and cribriform growth identified in whole-mount histological specimens.

Methods

A retrospective cohort of 291 patients with biopsy-confirmed PCa who underwent radical prostatectomy and pre-operative MRI was analyzed. Tumors were graded according to the 2019 ISUP classification, and cribriform growth presence and length were assessed on whole-mount histology. The oxygen consumption and supply based model using apparent diffusion coefficient and fractional blood volume was applied to estimate hypoxia level (HL) and hypoxia fraction (HF_DWI). Differences in HL and HF_DWI across Gleason patterns and between cribriform growth-positive and -negative tumors were assessed. Furthermore, a linear regression model was used to evaluate the association between cribriform growth and hypoxia after adjusting for confounders.

Results

Cribriform growth-positive tumors exhibited significantly higher HF_DWI values compared to cribriform growth-negative tumors (p < 0.001). Within individual cancer Grade Groups (GG), cribriform growth length was significantly associated with increased HF_DWI in pGG 3 tumors. HL_median values were highest in Gleason Pattern 3 (GP3) regions and lowest in cribriform growth (GP4Crib+) regions, with significant differences observed between GP3 vs. non-cribriform growth GP4 regions (p < 0.001) and GP3 vs. GP4Crib+ (p < 0.001), consistent with more hypoxia in GP4Crib+. Multivariable regression confirmed cribriform growth presence as an independent predictor of increased hypoxia fraction, even after adjusting for tumor volume and GG.

Conclusions

This study shows an association between cribriform growth and tumor hypoxia using an MRI-based biomarker. These findings provide further biological insights into the aggressive nature of cribriform growth architecture and highlight the potential clinical utility of non-invasive hypoxia quantification for risk stratification in prostate cancer management.

Clinical trial number

Not applicable.

Background

Prostate cancer (PCa) is one of the most frequently diagnosed malignancies among men, representing a highly heterogeneous clinical course, ranging from indolent tumors to aggressive, metastatic disease [1–4]. Therefore, precise risk stratification is crucial for guiding treatment decisions. Several histopathological biomarkers have been identified to predict prognosis, among which tumor hypoxia has been recognized as a key adverse prognostic factor [5, 6].

Hypoxia, characterized by reduced oxygen availability in the tumor microenvironment, plays a major role in cancer progression, treatment resistance, and poor clinical outcomes. The biology of hypoxic tumors have been extensively studied revealing increased genetic instability, enhanced angiogenesis, and resistance to both radiotherapy and chemotherapy, making them particularly difficult to treat [7–9]. In prostate cancer, hypoxia has been consistently associated with disease recurrence, metastasis, and overall mortality, securing its role as a well-established marker of tumor aggressiveness [5, 6, 9]. Hompland et al. developed an MRI-based hypoxia model, the consumption and supply-based hypoxia (CSH) model, which demonstrated strong correlation between the imaging-based hypoxia fraction (HF_DWI) and pimonidazole staining, a histological hypoxia marker [8]. This model was externally validated, further confirming the association between HF_DWI and tumor aggressiveness [10]. Additionally, a positive correlation was shown between HF_DWI and pathological Grade Groups (pGG), reinforcing its role in assessing tumor aggressiveness. It is the first imaging tool to map individual hypoxia levels from mild to severe, and thereby providing information not previously assessed [11].

In addition to hypoxia, cribriform growth is a relatively recent separated histopathological architecture that has emerged as a significant adverse prognostic factor in prostate cancer. First recognized as an independent predictor of poor outcomes in 2011 by Ickwoski et al. [12], cribriform growth architecture has been associated with higher rates of biochemical recurrence, metastasis and reduced survival [13–17]. However, despite its growing recognition, the biological mechanisms underlying its aggressive nature remain poorly understood.

Cribriform growth is characterized by its unique three-dimensional architecture, forming a serpentine structure where cells do not establish direct contact with the stromal microenvironment [18]. This structural organization differs from other prostate cancer patterns and has been hypothesized to create a distinct tumor microenvironment. The absence of direct stromal interaction may contribute to reduced vascularization and restricted oxygen diffusion, potentially leading to intratumoral hypoxia [7]. Chua et al. reported a weak correlation between cribriform growth and hypoxia based on intraprostatic measurements of pO2 levels [15]. However, the relationship between cribriform architecture and hypoxia remains poorly established. This is primarily due to the absence of clinically feasible methods to directly measure tumor hypoxia. The use of the CSH imaging model—a non-invasive, MRI-based hypoxia biomarker—offers a unique and promising opportunity to further investigate the association between cribriform and hypoxia. Therefore, to fill this knowledge gap, this study aims to associate hypoxia obtained from MR imaging with cribriform growth obtained from whole-mount histological specimens.

Methods

Patient cohort

This retrospective study included 291 men (Table 1) who underwent radical prostatectomy between January 2010 and December 2020 at the Netherlands Cancer Institute. All patients had biopsy-confirmed prostate cancer and pre-operative multiparametric MRI (mpMRI). The study was approved by the institutional review board (IRBd21-108), and informed consent was waived due to the retrospective nature of the analysis. Patients who had undergone prior transurethral resection of the prostate, had received neoadjuvant systemic therapy or radiotherapy, had incomplete or poor-quality MRI, or poor-quality histopathological specimens were excluded from the analysis. This cohort has been previously described [10, 19].

Table 1.

Patient demographics and characteristics of histopathological segmentations

Finding	Cohort
Patients’	N = 291
Age (y)*	67 (46–78)
PSA (ng/ml)*	8.3 (3-259)
Radical prostatectomy Tstage
2	168 (58%)
3	123 (42%)
Lesions	N = 319
Radical prostatectomy GG
1	45 (14%)
2	147 (46%)
3	76 (24%)
4	30 (9%)
5	21 (7%)
Cribriform growth presence on histology
Negative	198 (62%)
Positive	121 (38%)
Regions of cancer on histopathology on MRI	N = 465
GP3	242 (52%)
GP4Crib-	176 (38%)
GP4Crib+	47 (10%)

Note. –PSA: Prostate-specific antigen; GG: grade group; GP3: Gleason Pattern 3; GP4Crib-: Gleason Pattern 4 subtypes excluding cribriform growth; GP4Crib+: Gleason Pattern 4 subtype cribriform growth. *Data are medians (min – max)

MRI acquisition and data analysis

MRI examinations were performed using both 1.5T (Achieva [n = 3]) and 3T scanners (Achieva [n = 161], Achieva dStream [n = 103], and Ingenia [n = 21], Philips Healthcare, Best, the Netherlands). The protocol included axial T2-weighted (T2w) turbo spin-echo imaging (Repetition time = [2175–10233 ms], Echo time = [110–130 ms]) with a field of view (FOV) from 140 × 140 to 284 × 284 mm² and slice thickness of 2.5 to 4 mm) and diffusion-weighted imaging (DWI). 2D Single Shot Echo Planar Diffusion-weighted images were acquired with a FOV ranging from 180 × 180 mm² to 381 × 381 mm², slice thickness of 2.73 to 4 mm and in plane resolution of 0.56 × 0.56 to 1.2 × 1.2 mm². The acquisition b-values for DWI included 0, 200, 800 or 0, 50, 300, 800 s/mm², with some cases including an additional high b-value (1400–2000 s/mm²). Apparent diffusion coefficient (ADC) maps and fractional blood volume (fBV) maps were computed using an intravoxel incoherent motion (IVIM) model with a segmented fit approach1 [10, 20]. Detailed acquisition parameters and IVIM parameter value estimation were previously published [10].

Tumor delineations on MR images (T2w, ADC and high b-value DWI scans) were performed by two observers (M.D.C. and M.F.S., both 1–2 years’ reading experience) blinded to the pathological ground truth. If required, consensus was reached after discussion with an experienced radiologist (I.G.S. 13 years of experience).

Histological analysis and correlation to MRI

Histological evaluation of radical prostatectomy specimens was performed using whole-mount formalin-fixed sections. Gleason grading and sub-pattern identification, including the designation of cribriform growth, were performed according to the 2019 International Society of Urological Pathology (ISUP) consensus guidelines [21]. Tumors, when bigger than 3 mm, were segmented by an uro-pathologist (M.A.S.G, 10 years’ experience) using SlideScore software [https://www.slidescore.com/] and a 82X magnification. Additionally, the presence of Gleason Pattern 3 (GP3), non-cribriform Gleason Pattern 4 (GP4Crib-), and cribriform Gleason Pattern 4 (GP4Crib+) in every tumor was segmented separately. Intraductal carcinoma of the prostate was excluded from this analysis. Only invasive cribriform growth patterns, as defined by the 2019 ISUP criteria, were included in the GP4Crib+ category. These areas with different histological GP within the tumor were defined as regions. Each lesion containing cribriform growth pattern (GP4Crib+) was labeled as cribriform positive. Moreover, the GP4Crib+ maximum length on whole-mount radical prostatectomy (in mm) was annotated. Tumor and Gleason pattern histological segmentations were manually propagated to T2w MRI images by aligning anatomical landmarks such as the prostate capsule and benign hyperplasia nodules, to ensure accurate spatial alignment [19]. Segmentation uncertainty in MRI region analysis (GP3, GP4Crib- and GP4Crib+) primarily arose from manual co-registration of histopathology to T2-weighted MRI and the partial volume effect, particularly in smaller regions. To address these issues, an erosion process was applied to all segmented MRI regions using the ndimage.binary_erosion function (SciPy, version 1.11.1). This approach ensures that the extracted ADC and fBV values of the regions within the tumor are more representative of the actual underlying Gleason patterns, minimizing contamination from healthy tissue or other patterns at the region boundaries. This process removed voxels at the region boundaries to minimize contamination from surrounding tissue. Isotropic erosion was applied using a cylindrical structuring element in the XY plane of the ADC map, which is more susceptible to partial volume effects. Erosion was restricted to slices with adjacent upper and lower slices, avoiding edge-related uncertainties. The erosion radius was set to remove one acquisition voxel isotropically, ensuring the extracted features accurately represented the Gleason patterns within the tumor while reducing the impact of registration errors and improving alignment for subsequent analyses [19, 22].

Hypoxia level and hypoxia fraction calculation using the CSH model

The CSH model relies on the ADC and fBV maps which relate to oxygen consumption and oxygen supply respectively [8]. In the external validation [10], to compensate for acquisition and analysis differences, ADC and fBV distributions were median-scaled prior to model application. The model defines hypoxic levels (HL) for each pixel based on a linear relationship

1

where ADC₀=0.79 × 10⁻³ mm²/s and fBV₀=0.43 a.u. represent the intercepts optimized for the strongest correlation with pimonidazole as shown in Fig. 1 [8]. The HL threshold for a hypoxic pixel was set to 1, with decreasing HL values indicating more severe hypoxia Finally, the HF_DWI (Eq. 2) was calculated as the number of pixels with a HL value below 1 divided by the total amount of pixels in the lesion as shown in Fig. 1.

Fig. 1.

CSH model application workflow. Lesion is depicted in a purple contour. H&E: Hematoxylin and eosin staining, T2-w: T2 weighted image, ADC: Apparent diffusion coefficient map, fBV: fractional blood volume map, HL: Hypoxia level (pixel-level), HF_DWI: Hypoxia fraction (lesion-level)

2

In addition to lesion-level information, a more detailed region analysis was done by calculating the HL_median for GP3, GP4Crib- and GP4Crib+ regions.

Statistical analysis

HF_DWI differences between cribriform growth positive and negative tumors, as well as for tumors with cribriform growth areas below and above median size for each specific pGG category, were calculated using t-test (parametric) or Mann-Whitney tests (non-parametric, two-groups) depending on normality and homogeneity assessed by a Shapiro-Wilk test and Levene’s test, respectively. Significant differences in HLmedian among Gleason Patterns groups (GP3, GP4Crib- and GP4Crib+) were searched for using Kruskal Wallis testing (non-parametric, three groups). Moreover, differences between Gleason pattern groups were calculated using Dunn’s post hoc analysis with Bonferroni correction. Finally, the contribution of cribriform growth presence to HF_DWI was investigated by correcting for multiple predictors simultaneously, using a robust linear model (statsmodels.robust.robust_linear_model). The choice of a robust linear model was meant to overcome the presence of outliers and skewness of the data. The studied predictors were pGG, tumor volume, cribriform growth presence and cribriform growth length. All analyses were conducted using the Python programming language v3.11 (Python Software Foundation, https://www.python.org/).

Results

The cohort’s (n = 291) clinical and pathological characteristics are summarized in Table 1. The median age was 67 years (min-max: 46–78), and the median PSA was 8.3 ng/ml (min-max: 3-259). A total of 319 tumors were segmented and the pGG tumor distribution was as follows: 14% pGG 1, 46% pGG 2, 24% pGG 3, and 16% pGG 4–5. Cribriform growth was present in 38% of tumors on whole-mount specimens. Within this cohort, 592 regions of GP3 (n = 264), GP4Crib- (n = 238) and GP4Crib+ (n = 90) were segmented on whole-mount histological sections from radical prostatectomy specimens and successfully registered to T2w scans. Following the erosion process, 127 regions were removed since less than two voxels remained in the segmentation. A total of 465 Gleason pattern regions were included, from which 10% were GP4Crib+ (Table 1).

HF_DWI was higher in tumors with cribriform growth presence compared to those without (Fig. 2). The median HF_DWI in cribriform growth-negative lesions was 0.14 (IQR: 0.04–0.24) compared to 0.22 (IQR: 0.11–0.30) in cribriform growth-positive tumors (p < 0.001).

Fig. 2.

Boxplot HF_DWI distribution for lesions with and without cribriform growth on whole-mount specimens. A statistically significant difference (p < 0.001) between the two groups is denoted by the asterisks (***)

Within individual grade groups, no statistically significant differences in HF_DWI were found when comparing tumors with and without cribriform growth (pGG 2: p = 0.17, pGG 3: p = 0.14, pGG 4 : p = 0.18, pGG 5 : p = 0.92).

The median length of cribriform growth regions on whole-mount specimens varied across grade groups (Table 2). A median length of 2.5 mm was found in grade group 2 tumors. For grade group 3 and 4 tumors a median cribriform growth length of 4.6 and 4.5 mm, while 18 mm was found in grade group 5 tumors. When comparing HF_DWI values for tumors with cribriform growth length above and below median per grade group category, only grade group 3 tumors showed a significant difference (p = 0.01) (Fig. 3). For grade group 2 tumors, there appeared also an increase of HF_DWI when the cribriform growth length is larger, though not significant (p = 0.19). For higher grade groups (pGG 4 and 5), cribriform growth length did not influence the tumor HF_DWI (pGG 4 : p = 0.90 and pGG 5 : p = 0.71).

Table 2.

Cribriform growth presence and histological median max length (in mm) per pGG category

pGG	Tumors	Tumors with GP4Crib+	Median GP4Crib + max length (mm)
1	45	0	-
2	147	50	2.5
3	76	48	4.6
4	30	15	4.5
5	21	8	18

Note. – pGG: radical prostatectomy grade group; GP4Crib+: Gleason Pattern 4 subtype cribriform growth

Fig. 3.

Cribriform growth length effect on HF_DWI per grade group category. HF_DWI boxplots are shown for lesions with above-median (green) and below-median (orange) cribriform growth length across grade group categories. A statistically significant difference was observed in pGG 3 (p = 0.01). No significant differences were found in pGG 4 (p = 0.90) or pGG 5 (p = 0.71)

Analysis of hypoxia levels across specific Gleason patterns revealed a distinct gradient in tumor oxygenation, as measured by HL values. Among the patterns examined, Gleason pattern 3 showed the highest median HL values, indicating the least hypoxic tissue, followed by Gleason pattern 4 without cribriform growth. The lowest HL values, and thus the most hypoxic regions, were observed in areas of cribriform growth, as shown in Fig. 4. Statistically significant differences were observed between Gleason pattern 3 and GP4Crib- (p < 0.001) and between Gleason pattern 3 and GP4Crib+ (p < 0.001), indicating a decreasing trend in HL values from pattern 3 to cribriform growth pattern 4. While the median HL value was lower in GP4Crib+ than GP4Crib-, the difference did not reach statistical significance (p = 0.06).

Fig. 4.

HL_median distribution across Gleason patterns. Gleason pattern (GP3) regions had the highest HLmedian values, while cribriform growth (GP4Crib+) regions had the lowest. Statistically significant differences were observed between GP3 vs. GP4Crib- (p < 0.001) and GP3 vs. GP4Crib+ (p < 0.001). The comparison between Gleason pattern 4 non-cribriform growth (GP4Crib-) and GP4Crib+ was not statistically significant (p = 0.06)

Cribriform growth and its length remained the strongest significant predictors of HF_DWI, with the narrowest confidence intervals, even after adjusting for all variables (Table 3). A robust linear regression model confirmed that this association with hypoxia was independent of tumor volume and grade group. While pGG 3 and 5 were also significant, their wider confidence intervals indicated greater uncertainty. In contrast, tumor volume and pGG 4 were not significant predictors.

Table 3.

Robust linear model regression model of associations with hypoxia (HF_DWI)

Parameter	Beta	p-value	0.025 CI	0.975 CI
Intercept	0.067	< 0.001	0.047	0.087
pGG 3	0.060	0.025	0.008	0.112
pGG 4	0.063	0.107	-0.014	0.139
pGG 5	0.135	0.022	0.019	0.251
GP4Crib+ presence positive	0.067	< 0.001	0.047	0.087
GP4Crib+ median length	0.007	0.005	0.002	0.011
Tumor volume (cc)	0.002	0.299	-0.002	0.005

Note.- Reference categories for pGG and GP4Crib+ presence are pGG 2 and GP4Crib+ presence negative respectively

Two examples of hypoxia mapping using the CSH model are presented in Fig. 5. In Case A, a single lesion with an extensive cribriform growth area shows visibly lower HL values in the cribriform growth region (HL_median = 0.88 a.u.) compared to adjacent Gleason pattern 3 areas (HL_median = 1.32 a.u.). In Case B, two lesions are shown: the right peripheral zone lesion (HF_DWI = 0.31) contains cribriform growth with lower HL values (HL_median = 0.79 a.u.), while the left peripheral zone lesion (HF_DWI = 0.25) contains non-cribriform Gleason pattern 4 exhibiting higher HL values (HL_median = 1.31 a.u.). For both lesions, the ADC map show similar signal intensities across both the cribriform growth region for the right lesion and the GP4Crib- regions for the left lesion (median ADC = 0.6 × 10^-6 mm²/s).

Fig. 5.

Examples of hypoxia mapping using the CSH model in two prostate cancer cases. (A) A single lesion with a large cribriform growth component demonstrates markedly lower hypoxia level (HL) values in the cribriform region compared to surrounding Gleason pattern 3 areas. (B) A patient with two distinct lesions in the peripheral zone: the right-sided lesion contains cribriform growth and shows low HL values, whereas the left-sided lesion contains non-cribriform Gleason pattern 4 with higher HL values. Delineation color legend : Lesion – Purple, GP3 – Green, GP4Crib- – Orange and GP4Crib+ – White

Discussion

In this study, the relationship between cribriform growth and tumor hypoxia in prostate cancer using an imaging-based approach has been investigated. These results contribute to the growing body of literature supporting cribriform growth as an adverse prognostic factor and suggest that hypoxia may play a mechanistic role in its aggressive behavior. A key contribution of this study is the use of the CSH model, an MRI-based imaging model, to non-invasively assess tumor hypoxia. While the relationship between cribriform growth and hypoxia has been hypothesized and poorly established [15], this study is the first to demonstrate this association through imaging techniques in a large cohort of patients, providing novel evidence of this relation.

Our results indicate that tumors exhibiting cribriform growth architecture tend to have increased HF_DWI values, reinforcing prior evidence with oxygen electrodes that suggested a link between this Gleason sub-pattern and reduced oxygenation in the tumor microenvironment [15]. These findings provide further biological plausibility for the association between cribriform growth and adverse clinical outcomes in prostate cancer [13, 16].

The HF_DWI biomarker is defined as the ratio of hypoxic voxels (HL below 1) to the total number of pixels within the tumor. Our findings indicate that cribriform growth regions exhibit lower HL values, particularly when compared to pattern 3 regions, thus indicating higher HF_DWI of the lesion where this pattern is present. Although the median HL was lower in cribriform growth than in non-cribriform pattern 4 regions, the difference did not reach statistical significance (p = 0.06). However, the numerical difference between GP4Crib+ and GP4Crib−, when compared to GP4Crib− and Gleason pattern 3 values, suggest a biologically meaningful difference that may not reach statistical significance due to the smaller number of cribriform growth regions.

Given that prostate tumors are composed of a heterogeneous mixture of histological patterns, the contribution of cribriform growth to overall tumor hypoxia is influenced by its relative prevalence and size within each grade group category. In grade group 2 tumors, where Gleason pattern 3 predominates and Gleason pattern 4 constitutes a smaller proportion, cribriform growth is observed in only a subset of pattern 4 areas (50 out of 147 presented GP4Crib+ regions). As a result, the overall tumor hypoxia fraction is largely driven by the hypoxic characteristics of Gleason pattern 3, which exhibit significantly higher HL values. Consequently, while cribriform growth in grade group 2 tumors (median length: 2.5 mm) demonstrates a trend toward reducing tumor hypoxia, its contribution remains detectable but not statistically significant.

In grade group 3 tumors, where Gleason pattern 4 becomes the predominant pattern and cribriform growth prevalence increases (48 out of 76 tumors presented GP4Crib+ regions), the median cribriform growth length (4.5 mm) is associated with a significant increase in tumor hypoxia fraction. Clear statistical differences were observed between tumors with smaller and larger cribriform growth regions, highlighting the impact of cribriform growth architecture on tumor hypoxia dynamics.

In higher grade group categories (pGG 4 and 5), the effect of cribriform growth on tumor hypoxia fraction is less pronounced. This is likely due to the coexistence of cribriform growth with other Gleason patterns 4 subtypes (GP4Crib-), which also exhibit low HL values (not significantly different from the HL values found in cribriform growth regions). Additionally, our study focused specifically on cribriform architecture, and Gleason Pattern 5 was not individually included in the analysis. While cribriform and solid growth patterns both lack stromal contact, our study was not designed to compare these subtypes directly. Further work is needed to determine whether differences in hypoxia imaging exist between these morphologies.

Overall, the influence of cribriform growth architecture on tumor hypoxia is most pronounced in lower grade groups, particularly grade group 3 tumors. This effect is likely driven by two key factors. First, the increased prevalence and larger size of cribriform growth in grade group 3 compared to grade group 2, leading to statistically significant differences in grade group 3, despite an observable trend in grade group 2. Second, its coexistence with Gleason pattern 3 in these tumors, which exhibit significantly higher HL values, thus lowering the overall tumor HF_DWI. In more advanced disease stages, the interaction of multiple aggressive histological subtypes complicates the isolation of cribriform growth’s specific contribution to tumor hypoxia.

The association between tumor volume and HF_DWI was also evaluated. While larger tumors tend to be more hypoxic, cribriform growth remains an independent predictor of hypoxia, even after adjusting for tumor size. This suggests that cribriform growth contributes to a hypoxic microenvironment beyond what can be explained by tumor size alone. This finding reinforces the need to consider histopathological characteristics in conjunction with tumor volume when assessing prognosis and treatment response in prostate cancer.

The ability to detect cribriform growth non-invasively using imaging biomarkers such as CSH may have important clinical implications. In particular, it could enable pre-treatment identification of patients with cribriform growth, who may benefit from treatment intensification strategies or exclusion from active surveillance. This may be especially advantageous over biopsy-based methods, which are prone to under-sampling and may miss focal or small cribriform components. Imaging could therefore offer a more comprehensive approach to tumor characterization and patient selection for clinical trials. This added value is illustrated in Fig. 5, where hypoxia maps reveal focal areas of low HL values corresponding to cribriform growth that are not distinguishable on conventional MRI sequences such as ADC. In one example, two lesions within the same patient show similar ADC signal but markedly different HL values, with the cribriform growth-containing lesion displaying significantly lower HL values, thus higher HF_DWI. These findings suggest that hypoxia imaging may provide complementary information to standard imaging, allowing improved non-invasive identification of aggressive histological subtypes such as cribriform growth.

Despite these promising findings, this study has several limitations. The retrospective nature of the analysis introduces potential selection biases, and the absence of direct biological validation of hypoxia through molecular markers such as pimonidazole staining limits the ability to definitively confirm the HL threshold used to define hypoxic pixels. Another significant challenge in this study was the uncertainty in aligning histopathological delineated regions with their corresponding MRI regions during the co-registration process [19]. This misalignment could lead to contamination from nearby tissues (either other Gleason sub-patterns or healthy tissue), introducing errors in the extracted ADC and fBV values. This issue is particularly relevant for smaller cribriform growth regions, which are more prone to MRI partial volume effects. To mitigate these issues, an erosion process was applied to all Gleason pattern regions. Additionally, it should be noted that the DWI sequences used in this study were not specifically optimized for intravoxel incoherent motion modeling, and as such, the derived fBV values may be affected by noise. Lastly, we did not distinguish between large and small cribriform growth patterns. Emerging data suggest these subtypes may differ in their biological behavior, however both small and large cribriform patterns on prostate cancer biopsies are associated with similar worse outcomes [23].

The ability to quantify hypoxia through imaging before histopathological confirmation has significant clinical implications, as it could enable early risk stratification and potentially guide treatment decisions. Future research should focus on validating these findings in larger, prospective cohorts and integrating molecular hypoxia markers to strengthen the biological evidence of the association between cribriform growth and tumor hypoxia.

Conclusions

In conclusion, this study demonstrates an association between cribriform growth and tumor hypoxia using an MRI-based imaging biomarker (i.e. HF_DWI). Establishing this link reinforces the prognostic value of cribriform growth and provide further insights into its role in tumor progression. Finally, this study contributes to the broader understanding of cribriform growth in prostate cancer and may offer valuable clinical applications by providing hypoxia information by non-invasive imaging before histopathological confirmation, aiding in improved risk stratification and treatment planning.

Author contributions

All authors contributed to the study conception and design. Material preparation and data collection were performed by M.A.S. Guimaraes, M. Deręgowska-Cylke and M. Fernandez Salamanca. Analysis, figures and the first draft of the manuscript was done by M. Fernandez Salamanca and revised by P.J. van Houdt, I.G. Schoots, T. Hompland and H. Lyng. All previously mentioned authors plus E. Bekers, H.G. van der Poel, P.J. van Leeuwen and U.A. van der Heide commented on the second version of the manuscript. P.J. van Houdt, U.A. van der Heide and I.G. Schoots were also in charge of supervision.

Funding

The author(s) declare financial support was received for the research, authorship, and/or publication of this article. This project has been founded by a KWF grant. Project number: 12975.

Data availability

Data that support the findings of this study are not openly available due to reasons of sensitivity.

Declarations

Ethics approval

The studies involving humans were approved by the Netherlands Cancer Institute - IRBd21-108. The study was conducted in accordance with the relevant local and institutional guidelines and/or regulations including the Declaration of Helsinki.

Consent to participate

The ethics committee/institutional review board waived the requirement of written informed consent for participation from the participants or the participants’ legal guardians/next of kin because pursuant to national legislation prior to 25 May 2018 (Opt-out) and General hospital informed consent was considered.

Competing interests

The authors declare no competing interests.