Colorectal liver metastases contract centripetally with a response to chemotherapy

A histomorphologic study


  • The study was reviewed and approved by the Conjoint Health Research Ethics Board of the University of Calgary.



Recently, there has been considerable interest in neoadjuvant chemotherapy for colorectal liver metastases. However, there is little information that defines how much liver should be removed after a favorable response.


Liver metastases from 2 groups of patients were analyzed: 25 metastases were evaluated from a group that did not receive chemotherapy and 26 lesions were studied from patients who had received systemic chemotherapy before resection. All patients except for 1 had 5-fluorouracil (5-FU), leucovorin (LV), and irinotecan (CPT-11); 1 had 5-FU and LV alone. The average duration of chemotherapy was 2.9 ± 0.7 months. Separate assessments of the histopathologic features of the central and peripheral portions of each tumor were made. The pathologist was blinded to all clinical information.


All of the untreated metastases had well-circumscribed borders. Irregular borders were seen in 6 of the postchemotherapy lesions (26%), which was particularly prominent in lesions that had significantly contracted. After chemotherapy, discrete islands of viable tumor cells outside of the main tumor mass were seen in 4 patients, but all were close to the peripheral margin of the tumor mass. Viable tumor cells were more frequent in the periphery of metastases, regardless of chemotherapy exposure. Central necrosis was prominent in untreated metastases, but disappeared after chemotherapy. In lesions treated with chemotherapy, central fibrosis was greater compared with untreated lesions.


After a partial response to chemotherapy, liver metastases shrank in a generally concentric fashion. These findings support the practice of removing less liver after downsizing with chemotherapy. Cancer 2008. © 2007 American Cancer Society.

In recent years a number of potentially useful systemic agents for colorectal adenocarcinoma have been developed.1–5 Before this surge of development the options for patients with liver metastases from colorectal adenocarcinoma consisted of best supportive care, a much more limited armamentarium of chemotherapeutic agents, and resection in a minority of cases. If resection was feasible, then this represented the only real opportunity for long-term survival.6–9

With the advent of new chemotherapeutic agents, reports emerged in which some patients with unresectable liver metastases had reductions in their disease burden that were sufficient to enable complete resection of their remaining tumor.10–16 That is, tumor shrank sufficiently to enable complete removal of all gross disease, while preserving sufficient functional hepatic parenchyma. In such patients median survivals of 37 to 42 months10–12, 16 and 5-year survival rates of 35% to 40% have been reported.13, 14 This compares favorably with patients who obtained partial responses to chemotherapy, but were still considered to have unresectable disease (median survivals 21–22.2 months).10, 12

Recently, some investigators have administered preoperative chemotherapy to patients who have resectable liver metastases.17–19 The rationale for this approach is that preoperative chemotherapy enables better selection of patients for resection and subsequent courses of chemotherapy. That is, responsiveness to preoperative chemotherapy may identify individuals who are more likely to benefit from postoperative chemotherapy. Similarly, appearance of extrahepatic disease during preoperative chemotherapy identifies individuals who would not benefit from resection. Whereas there is a good theoretical rationale to administer neoadjuvant chemotherapy to patients with resectable liver metastases, this approach is investigational.20

One issue relevant to the utility of neoadjuvant chemotherapy for liver metastases is that it is unclear whether a favorable response to preoperative chemotherapy implies that it is possible to resect less hepatic parenchyma to completely eliminate remaining intrahepatic disease. That is, after a reduction in the size of a metastasis, is it possible to completely remove any residual tumor with a more limited liver resection? The answer to this question depends on the pattern of the response. We postulated that a concentric pattern of shrinkage occurs because cells at the tumor periphery are killed with great prejudice, resulting in a smaller tumor with no viable tumor cells distant from the main residual tumor mass. If that pattern were evident, then it would be justifiable to remove more limited amounts of liver parenchyma after involution of tumor (Fig. 1). The alternative pattern of response to chemotherapy is 1 in which cell death occurs randomly throughout the peripheral and inner portions of the tumor. If that response pattern occurred, then a partial response may result in multiple islands or nests of residual viable tumor throughout the entire area where the original (prechemotherapy) tumor had been. If this occurred, then it would be important to resect all segments of liver that were involved before chemotherapy.

Figure 1.

The clinical problem: After a response to chemotherapy, it is unknown whether (A) it is necessary to remove all segments initially involved with tumor or whether (B) it is safe to remove less liver (ie, the segments containing residual tumor).

The main objective of the study was to delineate the pattern of tumor regression, as this might help to define the optimal surgical approach after chemotherapy for liver metastases. For this reason, we evaluated the histomorphologic features of liver metastases after a response to chemotherapy, comparing those lesions to metastases that had not responded to chemotherapy and to metastases that had not been exposed to chemotherapy. Our findings that viable tumor tissue is rarely present farther than a few millimeters away from the main residual tumor mass supports the postulate that tumor shrinks in a concentric fashion and that a more limited hepatic resection can be safely performed.

Regional differences in the rate of chemotherapy-induced apoptosis or fibrosis may be a function of regional differences in vascularity, as antineoplastic agents may be delivered preferentially to areas that are better vascularized. Moreover, the fibrosis and the centripetal pattern of shrinkage that were apparent in the initial histomorphologic studies may be partly a vascular phenomenon. These regional differences may also be a function of the rate of proliferation of tumor cells, as rapidly proliferating cells are typically more vulnerable to chemotherapy. To evaluate these possibilities the secondary objectives of this study were to evaluate the patterns of the vasculature of the metastases and to evaluate the proliferative indices of tumor cells in each metastasis.



The study was reviewed and approved by the Conjoint Health Research Ethics Board of the University of Calgary. A description of the patients is summarized in Table 1. All patients underwent a hepatic resection for colorectal liver metastases between July 2001 and July 2004. One group of patients consisted of 20 patients who had not received chemotherapy before resection of their liver metastases. A total of 25 lesions were evaluated in this group.

Table 1. Clinical Features of the Study Groups
 No chemotherapyChemotherapyP
  • NS indicates not significant; N/A, not available; OR, operation.

  • *

    Indicates size immediately prior to OR in chemotherapy vs no chemotherapy groups.

  • Indicates size change prechemotherapy and postchemotherapy in the chemotherapy group.

No. of patients2021 
Age, y66 ± 1157 ± 9.006
No. of metastases
 No. of metastases per patient1.3 ± 0.55 (1–3)1.2 ± 0.51 (1–3)NS
Size of metastases, cm
 Initial size prechemotherapyN/A7.3 ± 4.3.061*
 Size immediately before OR6.3 ± 3.4*4.4 ± 3.8*<.0005

Systemic chemotherapy was administered before resection in 21 patients; 16 of these patients were participants in a phase 2 clinical trial of neoadjuvant chemotherapy for resectable liver metastases.20 All patients except for 1 had 5-fluorouracil (5-FU), leucovorin (LV), and irinotecan (CPT-11); 1 had 5-FU and LV alone. The average duration of chemotherapy was 2.9 ± 0.7 months (range, 0.5–5 months). Computed tomography (CT) scans were performed before and after chemotherapy, which enabled assessment of chemoresponsiveness. Eighteen patients (86%) had a response by RECIST criteria.21 This cohort was deliberately biased to include a large proportion of tumors that had responded to chemotherapy, and this response rate did not represent a consecutive group. A total of 26 lesions were evaluated in the chemotherapy group: 3 lesions that disappeared radiographically after chemotherapy; 19 lesions underwent a partial response; and 4 lesions were stable or progressed on chemotherapy. Surgery was typically performed within 4 to 6 weeks of completion of chemotherapy. In patients who had a response to chemotherapy, it was routine to remove all of the hepatic segments that were involved before initiation of chemotherapy, with at least 1–2 cm of gross tumor-free margin. Patients with positive margins were not included in the analysis.

Specimen Preparation

The gross appearance of each resected liver specimen was described by the receiving pathologist. The resection line was painted with India ink and the specimen was serially sectioned into slices measuring 1–2 cm. The number, distribution, and morphology of metastases in the specimen were recorded. Tumors were initially sampled based on 1 section per 1 cm of tumor mass, including adjacent normal liver parenchyma. One slide of tumor-free liver parenchyma was examined as well. These sections were embedded in paraffin and routinely processed. Subsequently, after reexamination appropriate sections from each tumor were selected for more detailed histologic analysis. Care was taken to ensure that the entire span of each tumor was evaluated at its maximal diameter, sampling the entire tumor if it was 4 cm in diameter or smaller. The larger lesions were examined by sampling 1 section per cm of tumor. Non-neoplastic liver was also sampled in each case, with sections taken away from the tumor. Representative areas from other sections were also examined.

Four slides of 4-μm serial sections were obtained from selected formalin-fixed paraffin blocks from each metastasis. Each section was mounted onto slides coated with polylysine-l. Immunohistochemical staining for CD31, CD34, and Ki67 was done in serial sections.

The pathologist examining the slides was completely blinded to each patient's treatment history. The parameters assessed for each tumor included a description of the tumor margin (sharply circumscribed or irregular), presence of calcifications, and presence of mucin pools. Other features (eg, fibrosis, necrosis, tumor cell viability, proliferative index, and vascular density) were assessed separately at the tumor periphery and at the tumor center. To accomplish this the diameter of a tumor was quartered; the outer quarters were evaluated separately from the inner (central) 2 quarters.

The degree of fibrosis was graded as follows: grade 1 (<30% of the area examined), grade 2 (31%–60%), and grade 3 (>60%). Necrosis was graded as follows: grade 1 (<25%), grade 2 (26%–50%), grade 3 (51%–75%), and grade 4 (>76%). Finally, the area occupied by viable tumor cells was estimated and graded as follows: grade 1 (<25%), grade 2 (26%–50%), grade 3 (51%–75%), and grade 4 (>76%).

The degree of vascularity was estimated by counting the number of CD34+ and CD31+ vessels per high power field (hpf; approximately × 200 magnification). An attempt was made to count 50 hpf in each case. However, it was not always possible, as some of the tumors were small and isolated metastases responded by subtotal necrosis. Subsequently, vascularity was graded. The grading for vessels staining for CD34 was as follows: grade 0 (0 vessels seen); grade 1 (1–20 vessels/hpf); grade 2 (21–40 vessels/hpf); grade 3 (>40 vessels/hpf). The grading for vessels staining for CD31 was as follows: grade 0 (0 vessels seen); grade 1 (1–40 vessels/hpf); grade 2 (41–80 vessels/hpf); grade 3 (>80 vessels/hpf). The grading system differed because there were generally more CD31+ vessels in liver metastases, although this was not a constant finding.

A proliferative index was derived by estimating the proportion of viable tumor cells that stained for Ki67. The index was assigned as follows: grade 0 (no cells staining positive for Ki67); grade 1 (1%–32% positive); grade 2 (33%–65% positive); grade 3 (66%–100% positive).

Statistical Analysis

All values are expressed as mean ± standard deviation unless otherwise specified. Statistical significance of differences between 2 means was tested by 2-tailed t-test for 2 independent samples. For comparison of grades (eg, fibrosis, necrosis, proliferative index), nonparametric tests were utilized; the Mann-Whitney test was employed to compare differences in medians of 2 independent groups; and the Wilcoxon Signed Rank test was used to compare differences in medians of paired variables. For categorical data the Pearson chi-square test was used to test the association of a factor with treatment (ie, chemotherapy or no chemotherapy). A P-value <.05 was considered significant a priori, but P-values less than .10 were reported.


Histomorphologic Features Associated With a Response to Chemotherapy

In patients treated with chemotherapy there was a significant reduction in size of tumors after chemotherapy (Table 1; P < .0005). Therefore, as one might expect, the average size of the lesions analyzed pathologically was larger in the group of patients who had not received chemotherapy, although this did not reach statistical significance (Table 1). Three lesions disappeared radiographically. In one, despite examination of 1-cm sections, there was no histopathologic evidence of residual tumor, including residual scarring. In 2 specimens evidence of residual tumor was seen. One of these consisted of hyalinized scar and the other consisted of some viable tumor cells surrounding a fibrotic center. The borders of each lesion were evaluated as sharply circumscribed or irregular (Fig. 2). All of the untreated metastases had well-circumscribed borders. Six of the postchemotherapy lesions (26%) had irregular borders, and all of these were reported in lesions that had undergone a significant size reduction.

Figure 2.

Comparison of some of the typical features of untreated liver metastases (A) and metastases that have contracted after chemotherapy (B-D). (A) Untreated metastases typically have a sharply defined border and central necrosis. (B) After a response to chemotherapy metastases often lose their sharply circumscribed peripheral border, which becomes irregular. Tentacles of viable tumor are occasionally visible, which appear separate from the main tumor mass. (C, D) The typically necrotic center of an untreated metastasis is replaced by fibrosis.

After chemotherapy, discrete islands of viable tumor cells outside of the main tumor mass were only seen in 4 patients, and none of these nests was more than 4 mm away from the peripheral margin of the main tumor mass. In addition, in 1 metastasis that had diminished in size after chemotherapy, the residual tumor consisted of 2 regions with distinct morphologies (Fig. 3). One region consisted of a large proportion of viable cells, and the other region was largely replaced by fibrosis and, to a lesser degree, necrosis.

Figure 3.

Regional differences in chemosensitivity. This postchemotherapy metastasis has 2 areas with distinctly different morphologies. Area A consists of a large focus of viable cells, whereas Area B consists of necrosed tumor that is largely replaced by fibrosis. As a result of regional differences in chemosensitivity, the boundaries of the lesion have become irregular. Note the viable tumor within the bile duct, which may represent a sanctuary from antineoplastic agents in Area C.

Viable tumor cells were more frequent in the periphery of metastases, whether they were treated with chemotherapy or not (Table 2). Chemotherapy-treated lesions that diminished in size did not differ in this respect compared with those that did not respond.

Table 2. Histomorphologic Features of Untreated Liver Metastases and Metastases After Chemotherapy
 No chemotherapyChemotherapyP
  • NS indicates not significant.

  • *

    Central vs peripheral (chemotherapy group).

  • Central vs peripheral (no chemotherapy group).

  • Chemotherapy vs no chemotherapy.

Viability, grade1.64 ± 0.572.48 ± 1.011.71 ± 0.81*2.67 ± 1.13*<.001*
Necrosis, grade2.28 ± 1.061.36 ± 0.761.54 ± 0.881.33 ± 0.70.003
Fibrosis, grade1.44 ± 0.511.48 ± 0.651.87 ± 0.74*1.46 ± 0.66*.035
Proliferative index2.04 ± 0.982.25 ± 0.681.54 ± 1.06*2.00 ± 0.78*.028*
CD31, grade0.55 ± 0.670.73 ± 0.830.63 ± 1.010.58 ± 0.77NS
CD34, grade1.16 ± 0.991.32 ± 0.901.44 ± 0.991.44 ± 1.08NS
 Well circumscribed2517.008
Sinusoidal vascular pattern

Central necrosis was prominent in untreated metastases, but became less prominent after treatment with chemotherapy (Table 2). After chemotherapy there was no significant difference in the degree of central or peripheral necrosis in lesions that responded and those that did not.

The degree of fibrosis was greatest in the central portion of lesions treated with chemotherapy. In lesions treated with chemotherapy the fibrosis was greater in the center than in the periphery, although this did not quite reach our predetermined level of significance (P = .051). Moreover, the degree of central fibrosis was greater in lesions treated with chemotherapy when compared with untreated lesions (P = .035). The mean central fibrosis grade was particularly high in metastases that shrank in response to chemotherapy (1.95 ± 0.78) versus those that did not respond (1.50 ± 0.58).

Vascularity of Metastases

Metastases were stained for the panendothelial cell markers CD31 and CD34. The number of vessels did not differ according to whether chemotherapy was administered, and vascularity did not appear to be related to the degree of response. There was little heterogeneity detected in the degree of vascularity in any given zone of each tumor. Moreover, in none of the groups could a difference in vascularity be seen between the central and peripheral components of each lesion. Therefore, it is unlikely that the pattern of vascularity markedly affects chemoresponsiveness of individual lesions or to particular regions in each metastasis.

The typical pattern of vascularity in metastases consisted of multiple small vessels distributed in a random fashion throughout the lesion (Fig. 4B); this is not unlike the angiogenic vascular pattern seen in most solid tumors at other sites. A sinusoidal pattern of vascularity (Fig. 4A) was observed in many lesions, where the tumor grew along hepatic sinuses with no significant neovascularization. These 2 vascular patterns have been described previously by others,22, 23 and the latter pattern is thought to represent parasitization of preexisting hepatic sinusoids. The sinusoidal pattern was seen in 61% of lesions exposed to chemotherapy and 72% of untreated lesions. This was not significantly different, suggesting that the appearance of a sinusoidal pattern of vascularity was not a treatment effect or association.

Figure 4.

Two angiogenic patterns in liver metastases; stained for CD34, × 20 magnification. (A) A ‘sinusoidal’ pattern of vascularity was seen in some metastases. (B) In others, a ‘portal’ pattern of vascularity is seen, which consists of multiple smaller blood vessels in a more random pattern and orientation.

Proliferative Index of Tumor Cells Distributed Throughout the Tumor

In metastases treated with chemotherapy, there was a significantly lower proportion of tumor cells that were proliferating at the center of the tumor when compared with the peripheral aspects of the tumor. This regional difference was not seen in untreated lesions, nor was it seen in lesions that did not respond to chemotherapy. When the proliferative indices of the inner portion of lesions treated with chemotherapy were compared with untreated lesions, no significant difference was observed (P = .12). When only lesions that responded were compared with untreated lesions, the proliferative indices in the inner region were less in the treated, responding lesions, although this was not statistically significant (1.53 ± 1.12 vs 2.04 ± 0.98, P = .12).


Neoadjuvant chemotherapy is becoming an increasingly more popular strategy for the treatment of liver metastases from colorectal cancer. This may be partly attributed to the higher response rates seen with contemporary antineoplastic regimens. However, with success a new conundrum has emerged: how much liver should be removed after a favorable response to chemotherapy? The answer to this question depends on the mechanism and pattern of tumor response. An apparent reduction in tumor size with a random pattern of cell death and persistence of viable tumor cells throughout the extent of where the original tumor had been would suggest that the entire segment involved with the prechemotherapy tumor should be removed. In practice, the resection strategy after neoadjuvant chemotherapy is based on the reduced dimensions of the metastases. Our data lend support to this practice. Because tumor shrinks in a concentric fashion, although there are occasional islands of viable tumor outside of the confines of the main tumor mass, it may be possible to remove less liver after a response to chemotherapy.

Before this study we speculated that tumor shrank in 1 of 2 ways: it either contracted in a piecemeal, random fashion, leaving viable cells throughout its original area, or it shrank in a concentric manner due to cell death that predominantly affected the periphery.24 Our data suggest an alternative, combined mechanism of response. Typically, in the absence of chemotherapy viable cells predominate in the periphery and central necrosis is prominent in metastases, presumably because of impaired perfusion to the core of the tumor. After a response to chemotherapy the distribution of viable cells remains unchanged, suggesting that the probability of cell death is the same in the peripheral and central portions of the tumor. Rather, with a clinical response to chemotherapy viable cells (and necrotic areas) are replaced by fibrosis, particularly toward the central portion of the tumor. Because responding lesions often lose their sharply circumscribed peripheral border and because discrete islands of residual viable tumor are not typically far from the main residual tumor, we believe that the chemotherapy-induced size reduction occurs primarily because the periphery is drawn toward the fibrosing center of the tumor (Fig. 5).

Figure 5.

(A) Two models of chemoresponse conceived a priori by the investigators. In the first model peripheral cells are preferentially killed by chemotherapy, leaving a central core of viable tumor. In the second model, residual disease is randomly distributed throughout the original tumor volume. (B) Model of chemoresponse derived from experimental data. Cell death is randomly distributed, due to regional variations in chemosensitivity. Elements of the center of the tumor are replaced by fibrosis, which draws remaining viable cells toward the center, effectively reducing tumor volume.

Little information is available concerning fibrosis within metastatic foci, but the appearance of fibrosis in other models of cell death may provide some insight into the role of fibrosis on the evolution of metastases. After myocardial infarction, hypocellular scars are seen at approximately 8 to 10 weeks.25 In the case of wound healing, dense hypocellular scars appear even sooner.25 In our series, fibrosis appeared to occur within 3 to 4 months of the initiation of chemotherapy. Given the delayed appearance of fibrosis after a myocardial infarction or after wound healing, we speculate that the majority of tumor death developed soon after the first few doses of chemotherapy, and fibrosis was initiated shortly after that.

If the replacement of dead tumor cells by fibrosis was the main mechanism responsible for reduction in tumor size, it remains unclear why no microscopically visible scar was detected in 1 of the lesions that completely responded to neoadjuvant chemotherapy. One explanation is that additional processes such as immune-mediated clearance or tissue remodeling may rid the region of any remaining scar tissue after tumor cells are completely eliminated. Given the appearance of a sinusoidal pattern of tumor vascularization, which may represent parasitization from hepatic sinusoids, tumor may have been replaced by regrowth of hepatocytes along the sinusoidal framework. Despite the considerable regenerative capacity of the liver, we consider this less likely. It is more likely that the residual scar was so small that it was missed by routine pathologic evaluation of the specimen, which consisted of examination of 1-cm liver slices. The observation by Benoist et al.26 that 80% of resected hepatic segments in which metastases had completely disappeared on CT contained viable tumor cells on careful pathologic examination supports this possibility. This is further supported by their finding that metastases that completely disappeared after chemotherapy have a recurrence rate of 74% if they were not resected.26

Whereas the centripetal implosion is a predominant process, one must be cognizant that there are regional differences in chemosensitivity within a single metastasis (Fig. 3). It is perhaps this phenomenon that resulted in the appearance of discrete nests of viable tumor cells outside of the main tumor mass. In this series, none of these viable tumor islands was ever found greater than 4 mm from the peripheral edge of the main tumor. In another series of liver resections for colorectal liver metastases, the size of the margin was not related to local recurrence rate or survival, although this was not assessed in the context of neoadjuvant chemotherapy.27 That is, as long as a histologically negative margin was achieved, an acceptably low rate of local recurrences was observed. Our findings of nests of residual viable tumor outside the main tumor mass suggests that a more conservative approach is required after neoadjuvant chemotherapy. Specifically, it would be prudent to attempt to achieve a 1-cm margin from the grossly visible residual metastasis, as this would result in histologically negative margins most of the time. Of course, more detailed studies, including outcomes studies, will be required to better evaluate this.

A lack of change in tumor cell viability in the central and peripheral aspects of metastases suggests that the rate of cell death is the same in both zones of tumor. On the other hand, the cells remaining in the center of residual tumor do change in their proliferative activity, as estimated by Ki67 staining. This may reflect the selection of nonproliferating clones after chemotherapy. If that is the case, because tumor cells residing in the peripheral aspects of the tumor did not have changes in proliferative index, it is possible that there is preferential cytotoxicity to centrally located tumor cells that could not be detected by the sensitivity afforded by counting viable cells.

Our findings are also analogous to observations seen in other tumor types. In breast cancer, for example, it is often possible to perform breast-conserving therapy after a favorable response to neoadjuvant chemotherapy, and this is associated with a low rate of locoregional recurrence.28, 29 Similarly, in the setting of rectal cancer, it is feasible to avoid an abdominoperineal resection in a greater proportion of patients after neoadjuvant chemotherapy, with survivals comparable to patients who did not receive neoadjuvant chemotherapy who underwent abdominoperineal resection.30 Therefore, as with other tumor models, with appropriate margins liver-sparing resections are possible, and these are unlikely to adversely affect outcomes.

The data should be interpreted with some caveats. This represents a relatively small series in which a detailed histomorphologic analysis was done. Larger series of chemotherapy-treated metastases should similarly be evaluated to define with greater certainty what constitutes a safe resection margin. All but 1 of the patients was treated with a combination of 5-FU, LV, and irinotecan. One patient was treated with 5-FU and LV, but was included because the patient was treated neoadjuvantly within the same time period and the histomorphologic characteristics of the resected metastasis were within the range exhibited in the other specimens. Although the decision to treat with neoadjuvant chemotherapy was not randomized, we are not aware of any major differences in patient or tumor characteristics between groups. The patients that did not receive neoadjuvant chemotherapy and those who were not in the trial were treated in the same institution by the same surgeons during the same time period. Patients receiving chemotherapy were younger, but there were no differences in sex, number, and size of metastases. Data from other chemotherapy regimens (eg, oxaliplatin-based regimens) should be evaluated to ensure that similar patterns of response are observed. With the advent of novel targeted therapies, particularly those targeting the tumor stroma (as opposed to the tumor cells themselves), further research on the pattern of shrinkage will be especially important. Future studies might also allow analyses comparing the dose and duration of chemotherapy to pathologic response. Another area for investigation would be to evaluate the relation between initial tumor size and response. The small sample of the present study precludes such analyses. Finally, whereas these data provide guidance for resection of metastases that have partially responded, the optimal management of metastases that have completely disappeared after chemotherapy is still poorly defined. Data from Elias et al.31 suggest that, in some instances of complete response, resection of the hepatic segments that had been involved may not be required, but such an expectant approach must be taken with considerable vigilance. Thus, further investigations on this line will be essential.

In summary, we have described a new mechanism of chemotherapy-induced tumor involution consisting of random cell death and replacement with fibrosis. Residual tumor is drawn in a centripetal direction, enabling resection of less liver in some instances. These findings lend support to the practice of not necessarily removing the entire segment of liver that was involved with prechemotherapy tumor.10–16 Because of the presence of occasional islands of viable tumor outside the confines of the main tumor mass, one should obtain generous resection margins if possible (preferably >1 cm). These data should be confirmed with larger series and with different systemic therapy regimens.